# Polylane changelog > New features, improvements and fixes in Polylane, release by release. Canonical: https://polylane.com/changelog/ · RSS: https://polylane.com/changelog/rss.xml ## GitLab October 31, 2026 Connect a GitLab group, on GitLab.com or self-managed. Polylane reads its code and links each project to what it deploys in the context graph. It watches pushes and pipelines, and opens merge requests from investigations, the same way it opens pull requests on GitHub. You can also sign in to Polylane with GitLab.com. See the [GitLab integration](https://polylane.com/integrations/gitlab/). ## Polylane for ChatGPT October 29, 2026 · Video: https://polylane.com/videos/for/polylane-for-chatgpt-1080p60.mp4 · Product Hunt: https://www.producthunt.com/products/polylane/launches/polylane-for-chatgpt Polylane is now an app in ChatGPT and Codex. Ask what changed, read logs and metrics, trace dependencies and investigate issues, all from your own code and infrastructure. Add the [Polylane plugin](https://polylane.com/for/chatgpt/), sign in, pick a workspace and ask. #### Views in the conversation Answers come back as views in the chat: issue cards, log lines, metric charts, recent changes and dependency graphs. Every resource in the workspace can be drawn as one context graph, with active issues marked, and an active-issues panel sits beside the conversation. Select anything in a view to ask about it next. #### Tools named for the job The plugin gives ChatGPT one tool per question you would ask: what changed, what it logged, what it reaches. ChatGPT picks the right one, and every answer links back to the evidence in Polylane. #### A guided first run The first time you ask, Polylane confirms the connection, picks your workspace, summarises what is connected and suggests a few first questions. ### Issue channels in Slack An investigation at the severity you choose now opens its own public Slack channel. Polylane posts milestones there, answers people in threads, names the channel after the issue and bookmarks the issue, the thread and the pull request. You choose who joins, from the Slack integration settings or from Slack itself. ### Improvements - Slack: Tables, timelines, stats and charts render as native Slack blocks - Slack: Polylane reads forwarded messages and files, and links a Slack member it recognises to their Polylane account - Slack: In Slack Connect channels, only members of your own team can ask Polylane for work - GitHub: A daily dependency scan files one issue per repository for critical and high severity upgrades - Pull requests: Reviewers on Polylane's pull requests follow fixed rules, and can be set per repository - Console: Save list filters as personal views on Issues, Escalations and Threads - Console: Chat about an escalation from its page, and the overview leads with what to do next - Google Cloud: Each Cloud Monitoring alert policy is assessed, and GCP projects are read and analysed like AWS accounts ### Fixes - MCP: The MCP grant is kept when a client retries a token refresh - Console: Dismissed notices stay closed - Console: Merged pull requests stay out of the Open filter - Console: Your written prompt is kept when an example question previews - Billing: One credit usage email instead of several, and invoices open with a current Stripe link - GitHub: Mentions are answered only in the workspace that knows the repository - Vercel: A 403 from Vercel is no longer read as zero projects ## Amp October 28, 2026 Connect Amp and it can write Polylane's fixes. Polylane investigates and writes the brief, then an Amp agent fixes the code and opens the pull request. Amp joins Devin, Cursor, Factory, Conductor and Capy. The token is checked with Amp before it's saved, and if a hosted agent fails, Polylane falls back to its own. See the [Amp integration](https://polylane.com/integrations/amp/). ## Amplitude October 28, 2026 Connect an Amplitude project with its API key and secret key. Agents can query its events, funnels and retention during an investigation. Polylane also checks the project's event volume once a day, so a drop in tracked events becomes an issue. See the [Amplitude integration](https://polylane.com/integrations/amplitude/). ## Braintrust October 28, 2026 Connect a Braintrust organization with an API key or service token, including organizations on the hosted EU data plane. Agents can search LLM traces, read a trace as a span tree, query logs and evals with SQL, and compare an experiment with its baseline. Polylane also watches each active Braintrust project for changes in errors, latency, tokens and cost. See the [Braintrust integration](https://polylane.com/integrations/braintrust/). ## New Relic October 26, 2026 Connect a New Relic account with one User key. Agents investigate with NRQL, traces, entities and alerts, and Polylane watches the account on its own. New Relic alert issues open Polylane issues, and the resources that report to the account are linked in the context graph. ## Polylane for Cloudflare October 23, 2026 · Video: https://polylane.com/videos/for/polylane-for-cloudflare-1080p60.mp4 · Product Hunt: https://www.producthunt.com/products/polylane/launches/polylane-for-cloudflare Polylane catches issues across your Cloudflare apps, follows the connections between Workers, databases, queues and storage, and finds the root cause using your code, logs and recent changes. When the fix is in your code, it opens a pull request for your review. See [Polylane for Cloudflare](https://polylane.com/for/cloudflare/). #### One read-only token Connect with a single Cloudflare API token. The create-token link pre-fills every permission Polylane needs. #### Every Worker and everything it binds to Workers, Durable Objects, D1, R2, KV and Queues land in one context graph, linked by their bindings, so an error in one Worker can be traced to the database or queue behind it. ### Improvements - Monitoring: Each connection gets its own signals: provider queries that watch one route, job or model call in your own field names, discovered daily - Monitoring: Saved queries and alert rules from your providers join each connection's checks - Escalations: Choose which criticalities alert straight away in Slack and by email, with a daily email of every open escalation - Integrations: Every integration has a Reconnect action, so you can renew a key without disconnecting - Pull requests: Choose which bot-opened pull requests, such as Dependabot or Renovate, are assessed automatically - Console: One menu filters every list and the pull request sidebar - Console: Workspaces can turn off videos on Polylane's pull requests - Slack: Channels are picked from one searchable list - Billing: Usage emails show what the credits paid for ### Fixes - Cloudflare: Pages bindings that point at deleted resources are skipped - Cloudflare: Client disconnects stay out of Durable Object error signals, and empty metric buckets read as absent instead of zero - Console: Issue and escalation pages stay loaded when you switch tabs - Vercel: Deployment list and detail reads work through the API tool - Integrations: A broken credential is emailed once, then weekly - Integrations: Axiom access is verified on connect, with an explanation when it fails - AWS: The AWS CLI snippet names the stack `PolylaneConnector` ## Google Cloud October 21, 2026 Every workspace can now connect a Google Cloud project. Setup is one continuous flow and keeps going in the background once you've granted access. Polylane investigates Cloud Monitoring alerts as they fire, and maps and monitors load balancers, data services and pipelines alongside the rest of your stack. See the [Google Cloud integration](https://polylane.com/integrations/gcp/). ## Polylane Fleet October 19, 2026 · Video: https://polylane.com/videos/polylane-fleet-dark-1080p60.mp4 Hard incidents have more than one lead. Any chat or investigation thread can now become a fleet: a lead agent and a team of member agents work the same production problem in parallel, each chasing a different lead. #### A shared board The agents coordinate on a shared git message board. They claim work, post evidence, and a second agent confirms a finding before it counts. People stay the authority over what the fleet concludes. #### Fleets in the console The thread sidebar has a Fleet section with the lead, every member, their lanes and the board. The threads list shows which threads ran as a fleet. Fleets are available on paid plans. ## incident.io October 18, 2026 Connect incident.io with one API key. Every new incident starts a Polylane investigation, and the findings are posted in the incident's Slack channel. Escalations Polylane can't fix page on-call through incident.io's own alert routing. Polylane sets up the incident.io workflow for you. See the [incident.io integration](https://polylane.com/integrations/incident-io/). ## LangSmith October 18, 2026 Connect a LangSmith workspace with one service key. Agents can investigate your LLM traces, and LangSmith alert rules open Polylane issues. Polylane also watches each tracing project for changes in errors, latency, cost and tokens. See the [LangSmith integration](https://polylane.com/integrations/langsmith/). ## Memory you can read and edit October 17, 2026 · Video: https://polylane.com/videos/polylane-memory-1080p60.mp4 · Blog: https://polylane.com/blog/how-to-build-a-memory-system-for-proactive-agents/ Polylane's memory is now a git repository of markdown files, one per workspace, with full history. Agents write to it as they work and it syncs at the end of every turn. You can read it, edit it and add to it from the console, the API and MCP. #### Browse it like code Open Memory in the console for a file tree and a markdown reader, with `MEMORY.md` first. Every memory has its own address, and edits happen in a slideover with Write and Preview tabs. #### One home for standing instructions Slack channel instructions and repository instructions for pull request reviews are now memory files, next to everything else Polylane knows about your workspace. #### Learns from how work ended A nightly update writes the day's note, removes entries that turned out wrong and rewrites `MEMORY.md`. It reads how work ended, such as merged or closed pull requests and decisions on issues, and writes a lesson only when a pattern repeats. ## Depot October 17, 2026 Connect a Depot organization with one token. Investigations can read container builds, GitHub Actions jobs and logs on Depot runners, and Depot CI runs, so a failing build is part of the evidence. Depot projects join the context graph with their own monitoring, and repositories that build on Depot are linked to their project from their CI and build files. See the [Depot integration](https://polylane.com/integrations/depot/). ## Neon October 17, 2026 Connect a Neon organization and Polylane adds your projects, branches, computes and databases to the context graph. Every service that reaches a Neon compute is linked to it, so a `DATABASE_URL` no longer ends at an unknown external host. Investigations get read-only SQL and query stats. Account checks and per-project consumption charts watch compute, transfer, storage headroom and failed operations. See the [Neon integration](https://polylane.com/integrations/neon/). ## Polylane Impact Analysis October 15, 2026 · Video: https://polylane.com/videos/polylane-impact-analysis-1080p60.mp4 · Product Hunt: https://www.producthunt.com/products/polylane/launches/polylane-impact-analysis · Blog: https://polylane.com/blog/how-we-prevent-slop-from-hitting-prod/ Catch issues before they reach production. Polylane assesses every pull request against the live infrastructure it touches: config, capacity, dependencies and telemetry. You get one comment with the evidence and a verdict on whether the change is safe to merge. See [impact analysis](https://polylane.com/use-cases/impact-analysis/). #### Missing observability, as a suggestion When a change ships code you could not debug in production, Polylane posts a separate comment proposing the logs, metrics or traces to add, linked to the line. When there is nothing worth adding, it stays quiet. #### An honest blast radius The assessment names the resources it could not assess instead of dropping them, and shows no number when it made no assessment. ### Improvements - Context graph: The context graph fits phone screens and responds to touch - Context graph: External services carry the health of the connected provider - Onboarding: Setup opens on the providers a scan of your domain found, and you choose Slack channels during onboarding - Escalations: Escalations are grouped by issue and repository, and show who handled them and when - Escalations: Unconfirmed escalations with no fix close after a week - Issues: An issue is resolved when its fix pull request is closed without merging - Console: A resizable file explorer for pull requests, and a Description tab that shows the walkthrough video on Polylane's own ### Fixes - Console: Lists stay steady during background syncs - Console: Metric cards no longer show invented zero values - Console: Issue state matches between the page and the filters - Console: Sessions survive transient errors, with an eight-hour idle timeout - Billing: Plan upgrades are charged when they happen, and video walkthroughs are no longer charged to your workspace - Notifications: A retried digest no longer emails every member again - Integrations: Better Stack connects with one global API token ## Capy October 14, 2026 Connect a Capy API key and Capy writes the fix. Polylane investigates the issue and writes the brief, then Capy opens the pull request in the project that holds the repository. Capy joins Devin, Cursor, Factory and Conductor as a coding agent Polylane can hand fixes to. When the repository isn't in a Capy project, you get an email and a Slack message saying so. ## Polylane for Vercel October 8, 2026 · Video: https://polylane.com/videos/for/polylane-for-vercel-1080p60.mp4 · Product Hunt: https://www.producthunt.com/products/polylane/launches/polylane-for-vercel Keep shipping while Polylane watches your Vercel apps. Install Polylane from the Vercel Marketplace and your projects, deployments, domains and logs land in one context graph. Agents catch regressions, find the root cause with evidence, and open a pull request with the fix for your review. See [Polylane for Vercel](https://polylane.com/for/vercel/). #### Start in Vercel, finish in Polylane An install started on the Marketplace carries through sign-up and workspace creation, and asks which workspace to use when you have more than one. #### Clearer token errors A Vercel token scoped to a single project is rejected with an explanation of the scope Polylane needs. ### Linked accounts Each member can link their Slack and GitHub accounts from a new Linked accounts tab. Polylane uses them for message attribution and commit authorship, even when the emails differ. ### PostHog error tracking Polylane receives PostHog error tracking events for new, reopened and spiking issues through a signed webhook, and watches the numbers on your PostHog dashboards. ### Improvements - Slack: Polylane speaks up in watched channels when someone reports a production symptom it can check, and posts what it found - Slack: Status updates are posted for background jobs that are still running - Console: Threads show Slack replies, background jobs and worker results - Console: Merged and closed pull requests are listed, and the sidebar has filter chips - GitHub: Add repository opens GitHub to choose more repositories, and Connect adds another GitHub account - Agents: Answers use charts and diagrams where they help - MCP: MCP consent defaults to read and write ### Fixes - Investigations: An investigation keeps going after a provider alert recovers - GitHub: GitHub connect can pick any account and its repositories - Console: Connect windows open on the click, so Safari no longer blocks them - Console: GitHub links show on review, merge, diff and file cards - Pull requests: Review comments are delivered again ## Introducing Polylane October 1, 2026 · Video: https://polylane.com/videos/polylane-topology-1080p60.mp4 · Product Hunt: https://www.producthunt.com/products/polylane/launches/polylane · Blog: https://polylane.com/blog/turn-your-app-into-a-context-graph/ · Blog: https://polylane.com/blog/proactive-agents/ Nobody should be on call. Polylane connects your code, infrastructure and observability data, investigates every issue and opens a pull request with the fix. When code can't fix it, you get the root cause and a recommendation. #### A setup thread that connects your stack New workspaces start in a chat where Polylane already knows your stack from a scan of your company's domain. It offers one connection at a time: clouds, tools, GitHub, Slack. After each one it takes a first look, showing what runs there and a 24-hour health chart, then suggests what to connect next. #### A pull request queue per repository Each repository has its own queue of open Polylane pull requests, with a limit you set per repository, so a backlog in one repository never holds up another. Admins can also have new Polylane pull requests opened as drafts. ### Improvements - Slack: Conversations stay responsive while long investigations run in the background - Slack: Messages about submitted pull requests show the severity - Console: Play review videos straight from thread cards - Console: Tables are simpler and lists load faster - Console: Usage is shown against your plan allowance ### Fixes - Context graph: Issue markers are easier to spot, and the legend explains its numbers - Cloudflare: The Create API token link pre-fills every permission Polylane needs - AWS: The connector's permissions are narrower, and it is denied reads of S3 objects and DynamoDB items - GitHub: Polylane never requests a review from a pull request's author - CLI: Reconnecting from the CLI works for every integration - Sentry: Throttled API calls are retried ## Convex production data October 1, 2026 Polylane could already see your Convex deployments, logs and metrics. Now agents can read the production data too: list a deployment's tables, read a table's newest documents, or run a one-off read-only query. No local MCP server needed: the same reads work from every investigation. See the [Convex integration](https://polylane.com/integrations/convex/). ## Video walkthroughs on pull requests September 25, 2026 · Video: https://polylane.com/videos/blog/autofix-walkthrough.mp4 · Blog: https://polylane.com/blog/nobody-reads-pull-request-descriptions-so-we-made-videos/ Pull requests Polylane opens now carry a short narrated, subtitled video in the description: the cause, the change, and how it was validated. Review the fix in a minute before you read the diff. ### Improvements - Investigations: Automatic investigations are paced against your open pull request queue and weekends, with separate severity thresholds for investigations and fixes - Notifications: Your team is told when pacing starts holding work back - Console: The Investigation tab on a pull request shows the live transcript - Console: Forecasts show history, the median path and an uncertainty band, in the console and in Slack - Console: Issues and Escalations are back as searchable pages, with counts in the sidebar - Onboarding: Connecting a cloud during onboarding happens in place, and sign-up ends with Slack - Kubernetes: New agents connect through a dedicated tunnel domain, and existing agents move over without reinstalling - Cloudflare: New checks for container start-rate refusals and D1 databases nearing their size limit - GitHub: Investigations pause cleanly when GitHub is disconnected, with a clear message ### Fixes - Context graph: Resource type counts cover every resource, not just the ones drawn - AWS: CloudWatch alarms with extended statistics update correctly - AWS: CloudTrail log files are read correctly - GitHub: CI alerts from a merge queue are keyed to the queued pull request - MCP: Credentials and internal error details are redacted from responses - Integrations: The ClickHouse connect URL is fixed ## PostHog September 23, 2026 Connect PostHog with OAuth, one project at a time. Investigations read your product analytics through read-only tools, so a spike in errors can be checked against what users actually did. Every connected project is checked daily for issues in its top events and its pinned dashboard trends, within the project's API budget. See the [PostHog integration](https://polylane.com/integrations/posthog/). ## Pull requests in the console September 18, 2026 The sidebar lists every pull request Polylane has open, with line counts, CI and review state, and filters for approved, failing CI and awaiting review. Each pull request has its own page, where you can merge or close it. ### Improvements - Slack: Agents post progress as one message they keep editing, which collapses to a one-line receipt when the reply lands - Slack: Notifications about the same work are threaded together and reply into the thread that asked - Console: Answers can include verdicts, stats, timelines and more chart types, including scatter, pie, heatmap and histogram - Context graph: Every resource has an environment, and monitoring of non-production environments can be switched on or off per account - Context graph: The graph draws progressively, with a progress bar - OpenTelemetry: OTLP log records and metrics are ingested - GitHub: `@polylane` works on pull requests Polylane did not open, and only workspace members can steer it - Console: Link a GitHub account from account settings if you signed up with Google or email - Console: A support page with docs, onboarding call booking and a shared Slack channel - Console: Trace spans open from the waterfall, and charts expand ### Fixes - Context graph: The graph loads on very large workspaces - Console: Cloud accounts can be disconnected from the Integrations pages - GitHub: Reconnect on an unavailable repository re-grants access and re-syncs it - Integrations: Logfire no longer recreates its alert channel every hour - MCP: The SSE stream stays alive, and bad credentials get proper errors - GitHub: Pull request titles are written in plain language ## Trigger.dev September 17, 2026 Connect Trigger.dev and failing background jobs become Polylane issues, one per error group per environment. Finished deployments show up as events, so a failure can be tied to the deploy that caused it. Agents get read-only tools over your runs, limited to what the API key allows. The integration settings show the webhook URL and take the signing secret. See the [Trigger.dev integration](https://polylane.com/integrations/triggerdev/). ## Commits from the teammate who asked September 11, 2026 Polylane's commits are now authored by the teammate who asked for the change, with Polylane as co-author. Platforms that only build commits from team members, like Vercel on private team repositories, now build Polylane's pull requests too. With no linked GitHub account it falls back to the workspace owner, then the Polylane bot, and a workspace can choose who Polylane commits as. ### See exactly what the agent read Every context item under a thread message now opens to show it as the agent read it: issues, cloud accounts, integrations, repositories, memories and resources, with credentials redacted. ### Improvements - GitHub: One investigation can open more than one pull request, such as a stopgap and the real fix, or a change in a second repository - GitHub: Polylane requests reviewers on its pull requests and keeps their state in sync with GitHub - GitHub: Untouched Polylane pull requests get a reminder after 7 days and are closed 7 days later - Integrations: Logfire alerts reach Polylane over a webhook - Console: Investigate again on an issue whose investigation has ended - Console: Polylane pull requests show Diff and Session tabs - Context graph: Hotspots count only open, confirmed issues, and the chat bar suggests questions - Slack: Links in Slack cards are inline, and buttons are kept for actions ### Fixes - Vercel: Team-scoped tokens work, and wrong-scope tokens get a specific message - GitHub: Review verdicts on Polylane's own pull requests reach the thread that owns them - GitHub: Lockfiles are kept when a dependency manifest changes - GitHub: A lost sandbox during submit is recovered instead of failing - MCP: Large results are flagged as truncated instead of cut off mid-JSON - GitHub: Connect an existing GitHub installation of your choice ## One investigation from issue to pull request September 4, 2026 · Blog: https://polylane.com/blog/sub-agents-are-just-wrong/ Every issue Polylane finds is now worked by a single agent. It reads the evidence with read-only tools over your telemetry and the context graph, and decides whether the issue is real. When the confirmed cause is a code defect in a connected repository, it writes the fix and opens the pull request in the same investigation. #### A clear outcome when there is no fix When a fix is out of reach, such as an unreachable repository, infrastructure code outside your connected repositories or a cause outside your systems, the investigation records the reason and a plain next step on the issue as an escalation. #### Your coding agent writes the fix With a coding agent connected, the confirmed fix goes to it. If a hosted agent rejects its credential, Polylane writes the fix itself. ### A second look at dismissed issues A dismissed issue that keeps recurring is looked at again, with a longer wait each time it is dismissed again. A dismissal is refused when its evidence does not rule out an incident, and a person's "no incident" stays final. ### Improvements - GitHub: Choose once per GitHub connection whether Polylane reviews pull requests, on by default - AWS: The connect dialog shows the generated CloudFormation template, with copy and download - AWS: `il-central-1` is available - AWS: RDS read replicas and RDS Proxy join the context graph, and API Gateway custom domains resolve to their targets - Context graph: Repositories are linked to the Kubernetes and Railway workloads running the images they build - Context graph: S3 buckets and RDS endpoints named in environment variables appear before the AWS account is connected - Cloudflare: Logpush jobs are linked to their sources and destinations, and gradual Durable Object memory climbs are detected - Railway: Agents can read variable names, never their values - Convex: Function metrics come from the log stream - Monitoring: A sustained deviation counts only when it moves the harmful way and goes beyond what that hour of the week usually reaches - Console: Paste or drop images into the chat composer - MCP: A consent screen shows the client, callback and scopes before you authorise it ### Fixes - GitHub: Polylane never closes a pull request it did not open - Grafana: Signals name the datasource they come from - Cloudflare: Durable Object namespaces are listed in full - AWS: Trust policy statement IDs are valid - Integrations: Sentry, GitHub and Slack connect errors show a friendly message - Slack: Only the person who installed the app gets the welcome message - Threads: A failed turn says so, with a clearer message when the model is overloaded ## Logfire September 4, 2026 Connect Pydantic Logfire and investigations can go from a failing route to the exact span that threw, with read-only SQL for everything else. Polylane watches each project's error rate, request volume and p95 latency like the other observability integrations, and names the service behind a rise in errors. See the [Logfire integration](https://polylane.com/integrations/logfire/). ## Grafana Cloud August 31, 2026 Connect Grafana Cloud, or self-hosted Grafana, with a stack URL and a service account token. Polylane adds a contact point and a catch-all notification route to the stack, so every existing alert rule, and every rule you create later, reaches it with no per-rule setup. Your own alert delivery stays as it is. Investigations can query dashboards, datasources, alert rules and annotations. If the contact point or route is removed, Polylane puts it back. See the [Grafana integration](https://polylane.com/integrations/grafana/). ## One AWS connection for every region August 28, 2026 Connecting AWS used to mean one region per connection. Now one connection covers any set of regions, or all of them, including regions you enable on the account later. Change the regions from the account at any time, with no CloudFormation changes. #### CloudTrail and alerts in every region New connections get a multi-region trail, and each selected region gets its own alert subscriptions. A CloudTrail event resyncs only the regions it touched. ### Blast radius on issues and threads The blast radius from pull request reviews now shows on issues and thread artifacts too, and agents can ask for it. It follows each relationship in the direction a fault travels, so it reaches the consumers two hops downstream that a fault degrades. #### Every connection, with its evidence A Connections tab on every resource lists its edges with the connected resource, the direction, how the link was discovered and the evidence behind it, such as the environment variable that created it. ### Improvements - AWS: CloudFront distributions are linked to their origins and WAF web ACLs - AWS: Security group rules become reachability edges in the context graph - AWS: Step Functions state machines are linked to every service they call - Cloudflare: R2 event notifications and custom domains join the context graph, and queue producers are linked to their queues - Cloudflare: Durable Object memory-limit resets are detected - Supabase: Auth, storage, realtime and REST services are linked to the database they front - Turso: Branches are linked to the database they forked from - ClickHouse: Services that share a data warehouse are linked - Slack: Tell the agent to go away and it mutes the thread until you mention it again - Slack: The Channels list shows every public channel, with a Join button for the ones Polylane is not in - Investigations: Set the lowest severity Polylane investigates on its own - GitHub: Polylane's commit messages and pull request titles follow Conventional Commits by default ### Fixes - AWS: Account IDs that are not 12 digits are rejected at connect - Cloudflare: Syncs stay within your account's API rate limit - Vercel: Telemetry ingestion works again - Mixpanel: Rate limits and timezone drift no longer erase monitoring coverage - Mixpanel: A missing raw export permission is named at connect - Slack: A Slack reply never fails silently - Console: Rate-limited metric reads say so - CLI: Accounts created through the CLI can sign in with a password ## ClickHouse Cloud August 24, 2026 Connect ClickHouse Cloud with an API key ID and secret, and your organisation's services join the context graph with their state, region and endpoints. Services that reach them through DNS records or environment variables are linked to them. A check flags unhealthy services, and investigations get ClickHouse tools for the service inventory, live metrics and the audit log. See the [ClickHouse integration](https://polylane.com/integrations/clickhouse/). ## Convex August 24, 2026 Connect a Convex team with a team access token and its projects and cloud deployments join the context graph, synced every 15 minutes. Services whose environment variables point at a `convex.cloud` or `convex.site` hostname are linked to the deployment behind it. Investigations can read a deployment's function logs. Each read uses a short-lived deploy key scoped to that read, which Polylane deletes afterwards. See the [Convex integration](https://polylane.com/integrations/convex/). ## Linear August 24, 2026 Connect Linear with an API key. It's the first issue tracker Polylane connects to: agents read your teams, projects and issues in any thread. From an investigation, Polylane can file a Linear issue with the evidence in its description, update its state, priority or assignee, and comment with findings. Issues filed from a thread link back to it. See the [Linear integration](https://polylane.com/integrations/linear/). ## Turso August 24, 2026 Connect Turso with a platform API token and your database groups and databases join the context graph, synced every 15 minutes. If the token reaches more than one organisation, you pick which one to connect. Checks watch your organisation's usage, and the token is validated and can be rotated like every other connection. See the [Turso integration](https://polylane.com/integrations/turso/). ## Resources linked to the providers that watch them August 21, 2026 Resources that report to Datadog, Sentry, Better Stack, Honeycomb or Axiom are now linked to that provider in the context graph, using the provider's own API as evidence. Before, the graph only knew about a provider when a resource carried its URL in an environment variable. #### Datadog and Sentry Hosts, databases, functions and queues that report through the Datadog agent or its AWS integration are linked to Datadog. Resources and repositories that send events to Sentry are linked to Sentry, refreshed every hour. #### Better Stack, Honeycomb and Axiom Resources watched by Better Stack monitors are linked to Better Stack, and resources whose names match a Honeycomb or Axiom dataset are linked to that provider. ### Severity on every issue The issues list has a severity column and filter, and each issue shows its severity in the header. Duplicates are hidden by default. ### Improvements - OpenStatus: Monitors are linked to the resources they probe - Coding agents: A connected coding agent writes Polylane's fixes, and Cursor agents get Polylane's MCP server so they can query Polylane while they work - MCP: Connect MCP servers that authenticate with custom HTTP headers - Slack: Charts in replies are attached as images - Slack: Long replies show live progress, with a link to follow along in the console - Axiom: The region is detected for you - Issues: An issue's lineage starts with the alert that opened it and names its provider - Agents: Answers with numbers come with a chart - Notifications: Informational emails are off by default, and routine fix pull requests no longer send one - Console: The new thread page puts the cursor in the composer ### Fixes - Supabase: Only the organisation you select is connected - PlanetScale: Only the organisation you select is connected - Sentry: Fixed failures that stopped Sentry and Better Stack connections working - GitHub: Adding repositories to an existing installation completes the connect - GitHub: Polylane's first pull request leaves symlinked instruction files alone - Integrations: Connecting a provider that is already installed no longer dead-ends - MCP: MCP servers get 60 seconds to connect - Console: Slack bot and unlinked user names show in threads, where they used to read System ## OpenStatus August 18, 2026 Connect OpenStatus with one API key. Polylane adds a webhook channel to your workspace, so every monitor failure, degradation and recovery reaches it as an alert. Monitors and status pages join the context graph, uptime checks run alongside the other providers, and investigations can read monitors, response logs, status pages, status reports and maintenances. See the [OpenStatus integration](https://polylane.com/integrations/openstatus/). ## Mixpanel August 17, 2026 Connect Mixpanel with a service account and Polylane finds every project the account can reach, so you pick the ones to connect. It's the first product analytics integration: an investigation can check whether signups dropped after a deploy. Agents can query event volumes, funnels, retention and saved Insights reports, and annotate charts. A daily check watches event volumes for tracking breakages. See the [Mixpanel integration](https://polylane.com/integrations/mixpanel/). ## Railway August 17, 2026 Connect a Railway workspace with OAuth or a workspace token and Polylane syncs its projects, environments, services, deployments, volumes and domains into the context graph. Pull request environments are left out. Deployments are watched through webhooks, service metrics are checked for issues, and investigations can read logs and use Railway tools. See the [Railway integration](https://polylane.com/integrations/railway/). ## A first pull request when you connect GitHub August 14, 2026 Connect GitHub and Polylane opens a pull request in one of your repositories within about a minute. It adds Polylane's instructions to `AGENTS.md`, `CLAUDE.md`, Copilot instructions and Cursor rules, and creates `AGENTS.md` when there are none, so the coding agents you already use know about Polylane. #### One per connection The pull request opens once, for a new GitHub connection, and Polylane skips archived and empty repositories when it picks where to open it. #### Follow it from the console The console shows the follow-up work after the first pull request as it happens, and you can cancel it. ### Ask the agent to watch something Ask the agent to keep an eye on a CI run or a release and it tells you what it is watching, then ends its turn. When the GitHub event lands, the thread wakes up with the result. Watches expire after a day by default. ### Improvements - Cloudflare: New connections use a read-only token, and the dialog lists every permission it asks for - GitHub: Polylane's pull requests carry a `severity:` label - GitHub: The production impact check completes on merge queue commits - Issues: Alert issues that stay quiet for a week resolve on their own, and reopen with their history if the alert fires again - Issues: Exceptions from error trackers start a notch lower in severity, and are raised when the evidence supports it - Escalations: Pull request and code references link to permalinks pinned to the commit - Console: A Quick answer button asks a long reply to wrap up - Console: Thinking and tool calls fold into one row per phase - Console: The appearance menu has a System option - Slack: The Slack agent speaks in the first person, with formatting converted for Slack - Slack: Images sent to the agent in Slack show in the console - Notifications: Emails have a new card layout ### Fixes - Console: Firefox loads every part of the console - Console: An interrupted reply shows a Try again notice - Console: Emoji shortcodes render in messages - Console: The page you were heading to survives signup and email verification - Cloudflare: Deleted Vectorize indexes leave the context graph - Agents: Internal IDs stay out of replies - Sign-in: Sessions renew correctly ## Talk to the agent while it works August 7, 2026 Sending a message while the agent was working used to cancel its turn and throw the work away. Now the composer stays live: your message is picked up at the agent's next step and shown in the transcript where the agent read it. #### One button to stop or send Stop and send share one button. Start typing and it becomes send; clear the text and it goes back to stop. `Esc` stops the turn. #### Answers that reach the agent Answers to the agent's questions and requests are tracked until the agent receives them, so one lost on a dropped connection shows as failed. Delivery receipts appear only when something goes wrong. ### A workspace from the first sign-in Signing up creates your workspace straight away, so you land in Polylane ready to connect your first provider. ### Improvements - Workspaces: Domain auto-join is off by default, with a switch in settings, and owners get an email on every join - Workspaces: A link into a workspace you are not in offers a request-to-join screen - Workspaces: The invite form is prefilled from the access request email - Security: An account needs a verified email before it can be used or join a workspace by domain - Slack: Tell the Slack agent how to behave in a channel and it saves those instructions - Slack: A Channels tab on the Slack integration lists the channels Polylane is in - Slack: Slack users without a Polylane account are invited to sign up - Context graph: Container registries, images and Terraform modules are resources in the context graph - Notifications: Every notification names its workspace and resource ### Fixes - Console: Sparse line charts draw across missing data points - Console: The in-thread Connect cloud card works for every provider - Slack: Replies no longer repeat their body and footer - Slack: An answer is posted even when no reply was drafted - Slack: Model reasoning stays out of replies - Logs: Capped log reads no longer show invented gaps in the log digest - Console: Counts use the right plural ## Conductor August 7, 2026 Connect Conductor and it can write Polylane's fixes. Polylane hands over the plan, a Conductor agent works on your repository in a cloud workspace, and the change comes back as a pull request. Conductor joins Devin, Cursor and Factory. The API key is checked with Conductor before it's saved, and you can make Conductor the workspace's default. See the [Conductor integration](https://polylane.com/integrations/conductor/). ## Modal August 4, 2026 Connect Modal with the token ID and secret Modal issues for CLI login, and your environments, apps, functions, sandboxes, volumes and queues join the context graph. A repository is linked to the Modal app its CI deploys, and Cloudflare DNS records that point at a `.modal.run` endpoint resolve to the app behind it. Investigations get Modal tools and a digest of each app's logs. Checks watch error log volume and queue depth against your account's history. See the [Modal integration](https://polylane.com/integrations/modal/). ## Supabase August 1, 2026 Connect a Supabase organisation with OAuth or a personal access token and Polylane adds every project to the context graph: databases and read replicas, edge functions, branches, storage buckets, auth, realtime and the REST API, synced every 15 minutes. Services that reach a project through a `supabase.co` hostname in their environment variables are linked to it, and Cloudflare Hyperdrive configs are linked to the database they point at. Each project gets dashboards, logs and agent tools, with checks on edge 5xx rate, Postgres errors, auth failures, function errors and request volume. See the [Supabase integration](https://polylane.com/integrations/supabase/). ## Follow an issue from check to pull request July 31, 2026 Every issue, investigation, fix and check now knows its neighbours. A lineage graph on the issue's Overview tab draws the chain from the check or change that raised it, through the investigation, to the fix and its pull request. #### Every step opens in place Click any card in the graph to open it in a panel. Duplicate issues point to the root issue that holds the investigation. #### In the API, CLI and MCP The same chain is available for any issue, check, thread, fix or pull request from the API, the CLI and Polylane's MCP server. ### CI failures become issues A failing GitHub Actions run on a connected repository now becomes a Polylane issue with the repository attached. Repeat failures of the same workflow fold into one issue, and a green run resolves it. Polylane reads the failed jobs, steps and check annotations, and a confirmed failure gets an investigation. ### The full analysis, one click from the comment Review comments open with a link to the full analysis on polylane.page: the complete write-up, interactive charts and full log excerpts. Polylane's own pull requests link to their whole run the same way. ### Improvements - Kubernetes: Connect a cluster through a read-only agent installed with Helm, so no kubeconfig leaves the cluster - Context graph: Services your code calls, like Stripe or Datadog, join the context graph from the URLs in environment variables - Context graph: Large context graphs open in seconds - Pull requests: The production impact check is on by default for every repository - Pull requests: A passing verdict cites the metrics or logs it rests on, and reviews list past issues on the same resources - Issues: Issues with traces show sample traces and request volume - Escalations: An Escalations page lists what needs a person, with a detail page for each - Threads: In threads with several people, Polylane knows who said what - Threads: Filter and sort threads by running, waiting and completed - Threads: Tell Polylane how you like to work and it remembers, for you alone - Workspaces: Admins can export a workspace, and a deleted workspace is held for 60 days before it's purged - Workspaces: A Privacy & data setting lets admins opt the workspace out of model training ### Fixes - Console: Sessions survive cold loads and refresh while you use them - Cloudflare: Syncing accounts with many Workers no longer floods errors - Monitoring: Metric buckets line up on a shared clock - Issues: Chains of duplicates point to one root issue - Slack: Messages from a newly installed Slack workspace are no longer dropped - Kubernetes: Kubernetes resources show short names - Console: Long inline code in issue text wraps - Threads: Generated thread names no longer contain tool calls ## Better Stack July 27, 2026 Connect Better Stack and its incidents become Polylane issues. Polylane sets up the outgoing webhook when you connect, and each incident is investigated with its cause and the monitor's response times by region around the time it fired. Agents can read your monitors, incidents, heartbeats, status pages and telemetry sources. Checks watch availability, incident volume, response time and monitors that are down. See the [Better Stack integration](https://polylane.com/integrations/betterstack/). ## Polylane in Slack July 24, 2026 Mention Polylane in Slack and it reacts straight away, shows that it's thinking and streams its reply as it writes. It appears in Slack's Agents tab with suggested prompts. #### Live progress Long turns post one status message and update it as each step completes. If a turn fails, Polylane says what happened and how to resume. #### Joins conversations it can help with In channels it belongs to, Polylane replies to messages it can help with even when nobody mentions it. Set a channel to mention-only to turn this off. #### Pull requests report back to the thread When a fix starts from a Slack thread, the thread gets the pull request link, CI failures and their fixes, review replies, and a reaction when it merges. ### Duplicate issues share one investigation Every new issue is checked against the open ones in your workspace. A confirmed duplicate is linked to the original and shares its investigation. Log lines are grouped by their shape, so an unfamiliar ID format no longer opens a new issue or reopens an old one. ### Improvements - Pull requests: Production impact verdicts post as a comment on the pull request, and Polylane replies when you comment on it - Pull requests: A changed verdict posts a fresh comment so GitHub notifies you, and a push that could not be reviewed is marked on the comment - Pull requests: Reviews check for deploy hazards: a rename and drop that old instances still read, a `CREATE INDEX` that blocks writes on a busy table, or an environment variable nothing provides - Pull requests: Anyone who can see the pull request can open the investigation behind a verdict, read-only, from the comment - Context graph: People and teams join the context graph from `CODEOWNERS`, deploy authors and owner tags, linked to what they own - Context graph: Hold a resource in the topology to trace its blast radius, and drag resources to rearrange it - Monitoring: Polylane writes checks for individual resources as well as accounts and integrations - Notifications: A weekly digest emails every member a summary of issues, pull request reviews, fixes and changes - Onboarding: Onboarding has a step to set up Polylane in your coding agent - Console: The fixes list has search, and its filters, sort and page are kept in the URL ### Fixes - Notifications: Notification emails reach the right recipients - Console: Email domain auto-join works for password sign-ups, and accepting an invite no longer shows a 404 - Notifications: Removing a cloud account sends one email, and removing an integration sends one too - Notifications: No fix email is sent when an investigation finds nothing to fix - Pull requests: Polylane's own pull requests get a production impact review - Context graph: The infrastructure graph is a top-to-bottom hierarchy again and opens at a readable zoom - Console: Editing a row updates it in place without reloading the table - Threads: Threads show their names in place of the agent's name ## Production impact review on every pull request July 17, 2026 · Video: https://polylane.com/videos/release-2026-07-17.mp4 Polylane now reviews every pull request for its impact on production. It follows the context graph from the repository to the resources the code ships to, checks the change against live topology and telemetry, and gives a verdict with the evidence and the affected resources. The verdict lands as the Polylane production impact check, which you can require in branch protection. #### Repositories linked to what they deploy Polylane reads Terraform, CloudFormation, Pulumi, CDK and deploy manifests such as `wrangler.jsonc`, `fly.toml`, `vercel.json` and `render.yaml` to link each repository to the resources it deploys. #### Your rules per repository Steer reviews for a repository with your own instructions. When a later push addresses a flagged concern, Polylane marks it resolved. ### Polylane's pull requests take feedback Comment on a pull request Polylane opened and it answers or pushes the change you asked for. Reply in the Slack thread where it was announced and it responds there. When CI fails on its branch, it reads the failed checks and pushes a fix. It follows your repository's `AGENTS.md`, `CLAUDE.md` and Cursor rules, and the commit message style in your history. ### Improvements - Vercel: Sandboxes, workflows, cron jobs, feature flags, AI Gateway, microfrontends, access groups, drains and container registries join the context graph - Vercel: Polylane follows the deployment serving production, even when a newer one exists - Cloudflare: Tunnels link to the zones, origin services and hosts they connect, and Hyperdrive links through the tunnel it uses - Cloudflare: Artifacts namespaces are synced with their metrics and checks - PlanetScale: Query Insights drive checks for query errors, p99 latency, latency spikes and query volume - Pull requests: Start a fix from a failing verdict with Fix with Polylane - Automations: Trigger and action filters pick cloud accounts, resources, repositories and Slack channels from dropdowns ### Fixes - Slack: Replies in Slack are formatted correctly - Repositories: Large repositories index without getting stuck syncing - Context graph: Large cloud accounts sync without hitting the size limit - Threads: A thread keeps showing progress when its agent restarts - Investigations: Model fallback also covers provider errors - Integrations: A failing credential check notifies once a day ## PlanetScale July 12, 2026 Connect a PlanetScale organisation with OAuth and Polylane adds your databases and branches to the context graph, for MySQL and Postgres alike. Services that reach a PlanetScale branch are linked to it by their connection host, so an error in a Worker or a container can be followed to the exact branch behind it. Investigations read Query Insights: query errors, latency and the windows where latency moved. Polylane never connects to your databases or holds database credentials. See the [PlanetScale integration](https://polylane.com/integrations/planetscale/). ## Polylane in your coding agent July 10, 2026 · Video: https://polylane.com/videos/release-2026-07-10.mp4 Your coding agent can now use the tools Polylane's own agents use. Connect Claude Code, Cursor, Codex or any MCP client to Polylane's MCP server and ask about production without leaving the editor. #### Ask from the editor Ask what errored in production for the service a pull request touches. The agent reads the context graph across your clouds, runs queries against your connected observability providers and searches your repositories. #### Scoped to your permissions Every tool is checked against your permissions, and only tools Polylane has listed for outside use are available. ### Alerts and detections are one list of issues Everything Polylane detects and every alert your providers forward now lands in one place as issues, with one lifecycle. The same problem firing again is one issue with an occurrence count. ### Improvements - Sentry: Sentry issues arrive with the exception attached: type, message, top stack frames and tags - Vercel: Paste a personal access token, validated as you paste it, to collect Vercel metrics and logs - AWS: After you launch the CloudFormation stack, the console waits for the account to report in and opens it - Monitoring: Turn monitoring off for a resource and Polylane keeps recording its changes without opening issues - Monitoring: Each check opens in a panel with its metric and recent evaluations - Issues: Issues have Metrics, Logs, Traces and Timeline tabs - Issues: Severity is set by the impact happening now ### Fixes - Vercel: A new deployment is recorded as one change - Console: Google sign-in returns to the right address - GitHub: Polylane pull requests closed without merging show as closed ## Factory July 4, 2026 Connect Factory and it can write Polylane's fixes. Polylane investigates and writes the brief, then Factory fixes the code and opens the pull request. Factory joins Devin and Cursor. Paste an API key from Factory's settings and it's validated before it's saved. Set Factory as the workspace's default for fixes. See the [Factory integration](https://polylane.com/integrations/factory/). ## Issue pages lead with the evidence July 3, 2026 Issue pages now open with what Polylane saw. Metric cards show each signal's value, its baseline and how far it moved, and a log summary groups the lines that matter by their shape. #### Blast radius Every issue shows the resources around the affected one in the context graph, so you can see what else is exposed before you open the investigation. ### Investigations survive deploys and restarts Investigations now run as durable workflows. A deploy or restart on our side no longer cuts one off mid-turn: it picks up where it stopped. ### Improvements - Cloudflare: New checks watch Hyperdrive queueing, Logpush failures, Durable Object errors and memory, Workflow retries, Workers AI, Vectorize, Browser Rendering and Tunnels - Cloudflare: Durable Object namespaces chart their SQL storage over time - Onboarding: Onboarding has steps to connect GitHub and Slack - Advisories: Each advisory says whether a setting change fixes it or it needs a rebuild or manual steps - Sentry: Webhooks from Sentry are verified with your integration's client secret - Investigations: When a model provider is at capacity, Polylane falls back to another model - AWS: X-Ray traces too large for one call come back as whole segments ### Fixes - Investigations: Stalled turns are swept automatically - Issues: Bot probe floods against paths like `wp-login.php` and `.env` no longer open issues - Sentry: Long Sentry alert titles are trimmed to a readable length ## Resource priority June 26, 2026 Polylane gives every resource a priority: Critical, Standard, Low or Minimal. It's worked out from the resource's connections, traffic and recent changes, and higher-priority resources are checked more often. #### Set it yourself Change a resource's priority from its page. A level you choose stays until you set the resource back to automatic. #### Reclassify an account Re-run the classification for every resource in a cloud account from the account's settings. ### Log summaries in investigations Investigations can summarise a large log result instead of reading every line. The summary gives the severity, the error count, recurring patterns and the most relevant lines, quoted verbatim. It covers Datadog, AWS CloudWatch, Vercel, Fly.io and Render logs. ### Improvements - Monitoring: Open a check evaluation to see the metrics, logs and traces it read - Cloudflare: The create-token link pre-fills every permission Polylane needs ## Every check evaluation is recorded June 19, 2026 Each time Polylane runs a check on a resource, a cloud account or an integration, the result is recorded. Resource pages list their recent evaluations, and the issues page of each cloud account and integration shows the evaluations of its checks. ### More of AWS in the context graph Kinesis streams, KMS keys, Secrets Manager secrets and service discovery services from your AWS accounts join the context graph. ### Improvements - Repositories: Polylane finds the GitHub repositories that deploy to a cloud account and links them to it - Onboarding: Connecting an account or integration ends on a page that suggests what to connect next ### Fixes - Context graph: An empty or failed read from a provider no longer removes resources from the context graph. A resource is removed once it has been missing for a day - Integrations: Calls to provider APIs are rate limited and shared between requests, so investigations and checks stay under provider limits ## More of Cloudflare in the context graph June 12, 2026 Polylane now syncs much more of your Cloudflare account into the context graph: zones, load balancers with their pools and monitors, health checks, Access applications, Tunnels, Spectrum apps, Stream live inputs, Images, Turnstile widgets, waiting rooms, certificate packs, Logpush jobs, and Workers routes and custom domains. #### DNS across providers A Cloudflare DNS record that points at a Vercel project, a Fly.io app, a Render service, an AWS load balancer or a CloudFront distribution becomes an edge in the context graph, so a hostname leads to the resource that serves it. ### Quieter in Slack threads Once Polylane is part of a Slack thread, it replies only to messages meant for it: questions, requests, answers to its own questions and corrections. Messages between teammates and acknowledgements such as "thanks" get no reply. ### Improvements - AWS: CloudFront distributions join the context graph - Workspaces: Custom model endpoints list the models they serve, so you pick one from a list - Workspaces: OAuth client secrets can be rotated without recreating the client ### Fixes - Slack: Links, channel names and user group mentions in Slack messages are read correctly ## Cursor June 7, 2026 Connect Cursor and it can write Polylane's fixes. Polylane investigates and writes the brief, then a Cursor cloud agent fixes the code and opens the pull request on your GitHub. Cursor joins Devin. Paste an API key from Cursor's settings and it's validated before it's saved. Set Cursor as the workspace's default for fixes. See the [Cursor integration](https://polylane.com/integrations/cursor/). ## Investigations read the whole provider API June 5, 2026 Investigations can now read any endpoint of the GitHub, Cloudflare, AWS, Vercel, Fly.io, Render, Datadog, Honeycomb, Axiom and Sentry APIs. When no dedicated tool covers what an investigation needs, it calls the provider's API directly with the access you already gave Polylane. #### Small responses API responses are trimmed before the investigation reads them. Link fields are dropped, long strings are cut and long lists are capped, so one large response leaves room for the rest of the evidence. ### Every fix has its own page Each fix opens on its own page, with an overview and a timeline of what happened from the investigation to the pull request. ### Improvements - Cloudflare: Investigations can search and read Workers traces - Cloudflare: Pipelines streams, pipelines and sinks join the context graph, with checks on decode errors, user errors and writes to sinks ## Kubernetes June 4, 2026 Connect a Kubernetes cluster and Polylane adds its namespaces, pods, deployments, stateful sets, daemon sets, jobs, services, ingresses, nodes, autoscalers and persistent volumes to the context graph, linked to each other. Checks watch for crash-looping pods, pods stuck in pending and unhealthy nodes. Investigations get read-only access to the cluster: they can list and describe resources, read their events and fetch pod logs. See the [Kubernetes integration](https://polylane.com/integrations/kubernetes/). ## Traces on issues May 29, 2026 Issues have a Traces tab. For AWS resources Polylane pulls X-Ray traces, and for Cloudflare Workers it pulls traces from Workers Observability. #### What the tab shows Trace counts, faults and latency over time, a breakdown by handler or operation, and sample error traces next to the slowest ones. ### Checks from your own dashboards Widgets on your CloudWatch dashboards and your saved CloudWatch Logs Insights queries become checks, and so do saved queries in Cloudflare Workers Observability. Polylane watches the signals you already chose to look at. ### Improvements - Sentry: Investigations can chart Sentry events over time ## Devin May 25, 2026 Connect Devin and it can write Polylane's fixes. Polylane investigates and writes the brief, then a Devin session fixes the code and opens the pull request on your GitHub. Create a service user in Devin's API settings and paste its token. Polylane detects the organization from the token before it's saved. Set Devin as the workspace's default for fixes when you connect, or change it later in settings. See the [Devin integration](https://polylane.com/integrations/devin/). ## Issues May 22, 2026 Polylane watches your connected cloud accounts and integrations and opens an issue when something looks unhealthy. Checks run every 10 minutes against AWS, Cloudflare, Vercel, Fly.io and Render accounts, and against Datadog, Sentry, Honeycomb, Axiom and GitHub. Polylane leans towards staying quiet: an issue opens only when the evidence is strong and specific. #### The evidence on one page Each issue has its own page with Polylane's reasoning, the metrics and logs behind it, a timeline and its properties. #### Every issue starts an investigation A new issue opens an investigation straight away, carrying the issue's summary and the charts that triggered it. ### Improvements - AWS: Kubernetes resources in your EKS clusters join the context graph, from namespaces and nodes to deployments, pods, services, ingresses and jobs - AWS: Investigations can get, list and describe Kubernetes resources on EKS, and read pod logs and events - Cloudflare: Investigations can list a Workflow's instances and their status - Context graph: Topology shows every connected cloud account in one context graph, filtered by provider or resource type - Fixes: A Fixes page lists every fix Polylane has worked on, with its pull request ## Fixes from a thread May 15, 2026 When a thread lands on a change worth making, ask Polylane to fix it. It writes a self-contained brief from what the thread found, makes the change in a sandbox and opens a pull request on your repository through the GitHub App. #### One change per pull request Each fix covers one logical change. Unrelated fixes from the same thread arrive as separate pull requests. ### Investigations follow your repository instructions Polylane reads `AGENTS.md`, `INVESTIGATIONS.md` and `CLAUDE.md` from the root of your connected repositories and follows them in every thread. ### Fork or rewind a thread Fork from any reply to start a new thread from that point, or revert a thread to an earlier message to drop everything after it. ## Polylane reads and posts in Slack May 8, 2026 Threads can work in your Slack workspace. Polylane lists the channels it can see, reads channel history and thread replies, posts messages and thread replies, reacts to messages and uploads files such as logs and images. #### Replies where the conversation is Polylane can read a Slack thread for context and reply inside that same thread, so what it finds lands next to the discussion it came from. ### More of Vercel in the context graph Vercel syncs now bring in each project's aliases, certificates, custom environments, DNS records, firewall configuration, log drains, storage, installed integrations and webhooks, so investigations see how a project is wired up beyond its deployments. ### Improvements - Pull requests: When Polylane opens a pull request with a fix, every member gets an email and your Slack channel gets a post with a link to it - Pull requests: A fix that fails is reported by email and in Slack too ## Cloudflare analytics in investigations May 1, 2026 Investigations can query your Cloudflare account through the GraphQL Analytics API. Polylane searches the schema for the dataset it needs, reads the fields that dataset offers, then runs the query, so it can pull request, error and latency figures for a Worker or zone in the middle of an investigation. ### Improvements - Vercel: Investigations can list a project's deployments and read any one of them in detail - Fly.io: Investigations can list an app's Machines - Alerts: Repeat firings of the same Datadog, Honeycomb, Axiom, Render or Sentry alert are rate-limited, and recoveries always come through