> ## Documentation Index
> Fetch the complete documentation index at: https://docs.aient.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Changelog
> Release notes for Aient covering new features, SDK updates, webhook changes, integrations, and bug fixes shipped each week to the platform.
## Updates
**Aient CLI 0.10.20 customer-agent messages.** We added `aient agent message --immediate` for cooperative delivery to an active task at a stable boundary. It never cancels or replaces that task. Reuse the stable operation ID when retrying an ambiguous result; the receipt reports whether Aient accepted, queued, started, or had already recorded the message.
**Aient CLI 0.10.17 and 0.10.18 safer long-running sandbox work.** Use `--detach --keep` when a sandbox command must outlive the initiating client, then inspect, attach to, wait for, or cancel the same execution by its stable ID. The CLI can also list, classify, and dry-run explicit runtime-service-to-repository mappings before applying them.
**Aient CLI 0.10.19 more resilient macOS project access.** Project access now survives a normal APFS remount while Aient continues to reject replaced directories. If an older definition needs repair, use `aient --profile NAME auth export --store-in-project` after confirming the host profile is healthy.
## Updates
**Aient CLI 0.10.16 project access.** On macOS, `aient auth login` now sets up bounded project access by default for a configured checkout. Aient keeps the refresh authority in Keychain and renews the project-local, short-lived access envelope in the background, so a verified checkout can continue working after the login terminal closes. `auth status`, `auth logout`, and `auth export --store-in-project` now make that lifecycle explicit.
**Selected nested repositories for sandbox work.** Sandbox runs and synchronisation can now transfer only the submodules a command needs. Use `--submodules=selected` with `--submodule PATH` to include an exact clean child repository, or `--submodules=gitlinks` when child contents are unnecessary. Aient rejects implicit or ambiguous submodule transfer before upload.
## Updates
**Customer MCP connections.** Organisation owners and administrators can now attach a trusted remote MCP server to one Aient environment, review its discovered tools, approve a bounded subset, and make it available to new Sparkles tasks. Connections keep credentials write-only and fail closed when their approval or configuration changes.
**Aient CLI 0.10.7.** The CLI now gives remote workspace synchronisation clearer path-ownership controls. You can preserve or reset named generated directories without changing local files, while incompatible server capabilities or competing workspace layouts fail safely before upload.
**OpenTelemetry quickstarts for more runtimes.** New setup guides cover standard OTLP/HTTP instrumentation for Python, Java, .NET, Ruby, PHP, Go, and Elixir, including the deployment metadata Aient uses to correlate traces with a release.
## Updates
**Aient CLI 0.9.1 for invited preview.** The CLI now uses negotiated atomic activation when it synchronises a workspace to a retained sandbox. If another workspace layout owns the target, it returns actionable contention guidance instead of dispatching or replaying against ambiguous state.
**Browser SDK click and identity lifecycle.** The Browser SDK can record one point span for each eligible user click. Apps can now identify, clear, and read their current correlation identity, while server-authenticated identity remains authoritative.
**Published posts are easier to share.** Aient blog posts now include copy-link, email, and X share actions, along with responsive author profiles.
## Updates
**Aient CLI 0.8.2 for invited preview.** Retained sandbox `status`, `wait`, and `logs` now use the environment that owns the sandbox. Owners and administrators can also set encrypted environment secrets from the CLI; values are write-only and never returned. Workspace synchronisation now treats `--timeout` as an inactivity limit while bytes are moving, so healthy large uploads can take longer than the default five minutes. A retry starts a fresh delta or snapshot.
**Safer customer sandbox control.** Teams can choose a product-governed `small`, `medium`, or `large` compute class for an environment-backed sandbox. Workspace synchronisation can now start from all eligible non-Git files and explicitly exclude paths such as build output and `.env` files.
## Updates
**Reopen resolved problems through MCP.** Assistants with MCP write scope can now reopen a resolved problem when new production evidence changes the conclusion, while preserving its existing evidence and remediation history.
**Clearer metered-usage billing.** Billing Settings now distinguishes recorded usage for the current period, estimated usage that has not yet reached an invoice, and provider invoices and payments. Interim invoices can be issued when outstanding usage reaches the subscription threshold.
## Updates
**GitHub App connection checks.** Aient now verifies the installer and ties each GitHub App installation to the selected organisation before linking it. If an installation is already linked elsewhere, Aient explains the next step to resolve it.
**MCP sign-in continuation.** MCP OAuth onboarding now carries a genuine sign-in or MFA flow back to the authorisation step while keeping ordinary visitors behind the launch gate.
## Fixes
* The browser SDK now ignores Firefox Reader Mode probe errors, keeping non-application noise out of your telemetry.
## New features
**Remediation evidence packet.** Problem case files, MCP `get_problem`, and remediation prompts now share a current remediation snapshot with PR bindings, verification evidence, prior handoff context, and exact artifact handles. Agents and humans start from the same case evidence instead of re-reading raw thread history.
**Public blog and feature pages.** Aient now has a public blog with static MDX, Mermaid diagrams, feeds, and SEO metadata, plus dedicated feature detail pages with product screenshots from the homepage.
**MCP setup in settings.** Teams can revisit MCP connection setup from settings, with shared configuration guidance across onboarding and signed-in setup surfaces.
## Updates
**Problem view readability.** Problem pages now lead with the case document and keep raw inspectors collapsed by default, while retaining raw details for debugging.
**Remediation review and verification loop.** Aient now tracks PR status, approval comments, delivery waits, deploy evidence, verification confidence, and human questions more consistently inside the active remediation attempt.
**Human answer continuation.** Answers to remediation questions now wake the same root agent task and stay in the same conversation lane, reducing lost context after a human decision.
## Fixes
* Public MCP and x402 thread tools now fail closed before reading private agent-designer replay threads or their artifacts.
* Source-map uploads and Restate ingress recover more reliably from transient network failures.
* Telemetry ingest now applies smaller ClickHouse batches to reduce pressure during high-volume periods.
* Resolved problems and terminal remediation PR feedback are less likely to reopen stale work or show old closure states.
* Agent remediation edits and workspace paths now preserve safer path boundaries and avoid exposing backing worktree details.
## New features
**MCP AWS Firehose drain management.** Authorized MCP clients can now list, create, rotate, and revoke AWS Firehose drains for an environment, giving agentic setup flows the same lifecycle controls as the dashboard.
**Source-map deployment tracking.** Source-map uploads now record deployment SHAs and compress artifacts on upload, giving stack trace resolution a more durable link between running code and retained source maps.
**Remediation run history.** Agent remediation runs now persist runtime events into thread records, so problem threads keep clearer evidence about what the agent did, where it got blocked, and how it recovered.
## Updates
**Remediation PR feedback loop.** Aient now binds GitHub bot reviews, inline review comments, PR status snapshots, branch evidence, and review-triggered waits into remediation threads more consistently, so agents can keep fixing PRs without losing current context.
**Remediation sandbox reliability.** Sandbox lifecycle checks now expose contract mismatches and readiness failures as explicit telemetry, improving recovery when a workspace resumes, hibernates, or reports an unexpected state.
**Source-map retention.** Retention now keys off deployment SHAs, service configuration, canonical commit SHAs, and safer object-delete paths, reducing accidental artifact loss and keeping symbolication evidence available longer.
## Fixes
* Early-access request submissions now throttle by requester identity, reducing duplicate requests without blocking unrelated teams.
* Billing webhooks now replay duplicate Stripe subscription and invoice events safely.
* MCP setup now tolerates placeholder organisation selectors while preserving authenticated tenant scope.
* Remediation agents recover more reliably from provider finish errors, content-filter interruptions, sandbox loss, and compacted context.
* Remediation PR delivery now preserves assigned branches, pushed head SHAs, active obligations, and already-advanced completion states more accurately.
* Source-map retention no longer times out on large keep-set queries or fails on MinIO delete checksum requirements.
## New features
**Public security page.** Aient now has a public security and trust page with responsible disclosure details, contact paths, and a published `security.txt`, giving teams a clearer place to review Aient's security posture before rollout.
**Teams inbound replies.** Microsoft Teams messages now flow through the same channel-turn ingestion path as other collaboration surfaces, so Teams replies can carry problem and remediation context back into Aient more reliably.
**Request access form.** Public Aient pages now route early-access teams to a first-party request flow, keeping rollout requests on-site and making the next step clearer for teams evaluating the platform.
## Updates
**Remediation status reliability.** Remediation threads now preserve the active source of truth when older blockers, review facts, or workspace epochs arrive late, keeping problem status, PR context, and verification history aligned.
**PR review context.** Agent-authored remediation work now hydrates more source context from GitHub comments and review events before applying corrections, improving follow-up quality on requested changes.
**Telemetry ingest health.** Vercel Drain and AWS Firehose ingest paths now report receiver health more consistently, making telemetry setup and troubleshooting easier to verify.
## Fixes
* Onboarding no longer shows a stale URL state after returning from OAuth.
* Service selection now stays in a clear loading state during long-running discovery instead of presenting a premature error.
* Slack workspace selection copy is clearer during onboarding.
* Guarded GitHub PR review failures are now surfaced in remediation threads instead of failing silently.
* Thread workspaces remain visible across remediation epoch changes.
## New features
**Vercel Drain ingestion.** Teams can now connect Vercel custom-endpoint Drains for logs and traces from **Settings → Environments → Vercel Drains**. Aient verifies Vercel signatures, stamps telemetry to the selected organisation and environment, and sends it through the same problem detection and remediation pipeline as SDK ingest.
**Vercel integration guide.** The docs now include a Vercel Drain setup guide with the Aient settings flow, Vercel destination steps, signature verification guidance, and what to expect once telemetry starts arriving.
**AWS CloudWatch Logs ingestion.** Teams can now connect CloudWatch log groups to Aient through Amazon Data Firehose from **Settings -> Environments -> Drains**. Aient accepts the standard CloudWatch Logs payload, verifies the copy-once Firehose access key, and turns error-class AWS logs into the same problem detection and remediation flow as other telemetry.
**AWS Firehose integration guide.** The docs now include an AWS setup guide with Firehose destination settings, CloudWatch Logs subscription filters, IAM role guidance, Terraform examples, verification steps, and common troubleshooting paths.
**Remediation PR review feedback.** GitHub review feedback on agent-authored remediation pull requests now routes back into the remediation thread, including requested-changes and dismissed-review states, so agents can react to review context without losing the PR binding.
## Updates
**Server-side source maps for agents.** Aient's instrumentation guidance now treats `@aient/sourcemaps` uploads as required for transpiled or bundled Node, Bun, and Deno services, giving server-side stack traces the same symbolication path as browser and Next.js bundles.
**Vercel observability page.** Aient now has a public Vercel integration page that explains the Drain-based path from Vercel logs and traces to problem detection, root cause analysis, and reviewed fix pull requests.
**Problem lifecycle reliability.** Problem transitions and projections now preserve monotonic lifecycle state when delayed or replayed events arrive, reducing stale status regressions in problem details, remediation threads, and downstream exports.
## Fixes
* Vercel Drain management now uses tighter admin controls, safer rotation and revoke behavior, and retry-aware handling for upstream ingest failures.
* GitHub identity-link notices in remediation threads now name the GitHub actor that needs linking.
* Source-map bulk upload signing now retries transient signing failures before blocking upload batches.
* Remediation PR status projections now keep verified and merged PR states visible on problem details and case files.
## New features
**Public agent setup page.** Aient now has a public `/agents` entry point for teams and agents that need setup paths, documentation links, and agent-commerce context before signing in.
**Pricing and comparison pages.** Aient now has public `/pricing` and `/compare` pages that explain the base-plus-usage model, budget controls, and how Aient differs from other AI SRE and autonomous remediation tools.
**Retry failed remediations.** Failed remediation attempts now expose structured failure evidence and retry actions from the dashboard and MCP tools, so teams can restart eligible work without losing the original problem context.
**Problem issue-kind labels.** Problem lists, details, routing context, and agent exports now show whether a problem looks like application code, infrastructure, database, or another issue type, helping teams send work to the right owner faster.
**MCP onboarding activation tools.** Agents connected through MCP can now read activation status, submit supported onboarding steps, and retry repository discovery when the token has the right scope.
**MCP service-to-repository linking.** Authorized MCP clients can now connect observed telemetry services to the repository and path Aient should use for source maps and remediation context.
**Telemetry-aware billing setup.** Aient now prompts billing setup after onboarding and first telemetry, keeping payment setup out of the way until the platform has real signal to protect.
**Teams as a collaboration channel.** Teams can now choose Microsoft Teams during onboarding, route problem notifications to Teams destinations, and manage collaboration delivery with the same readiness checks used for Slack.
**Official MCP Registry listing.** Aient's hosted MCP server is now listed as `ai.aient/mcp` in the official MCP Registry, giving coding agents a standard discovery path for Aient context and remediation tools.
**Confirmed problem resolution.** Dashboard owners can now confirm a pending agent-proposed resolution from the problem action flow, keeping closure explicit while preserving the agent's evidence trail.
## Updates
**Source-map evidence in problem cases.** Problem case files now include resolved source-map stack details and call-chain evidence, giving coding agents better context before they prepare a fix.
**Richer problem link previews.** Shared problem links now unfurl with the real problem title in supported clients, making incident context easier to recognize before opening Aient.
**Remediation workspace reliability.** Aient now handles workspace materialization failures as explicit states and authenticates GitHub operations more consistently when preparing remediation pull requests.
**Problem list scalability.** Large problem, label, dashboard, service-health, and scanner reads now split oversized filters into bounded requests, so high-volume organisations keep loading reliably as their history grows.
**Remediation PR readiness.** Aient now evaluates GitHub commit status contexts alongside check runs before treating an agent-authored remediation PR as ready for review.
**GitHub App key compatibility.** Aient now accepts the PKCS#1 private keys GitHub Apps provide and normalizes them for JWT signing, reducing setup failures for GitHub-backed remediation.
**MCP setup scopes.** MCP setup now shows read and write scope controls, with generated configuration that requests write access when remediation control is needed.
**Browser error instrumentation hardening.** The browser SDK now tolerates null or missing runtime error messages while filtering hydration mismatch noise.
## Fixes
* Remediation attempts can now record already-fixed, no-fix, and awaiting-human outcomes without turning them into false "no PR created" failures.
* Problem routing and verification now avoid more stale retry loops, reducing stuck or misleading remediation states.
* Problem action controls now have more consistent parity across problem surfaces.
* Agent execution budgets now account for tool time more accurately, reducing false retry exhaustion.
* GitHub authentication is now applied more consistently when remediation agents publish changes.
* Telemetry verification and anomaly debugging now derive tenant scope from authenticated access instead of caller-supplied identifiers.
* Agentic GitHub onboarding now verifies installer control before linking an installation to an onboarding session.
* Teams destination deletion now stays scoped to Teams routes.
* The Problems page no longer records URL-state hook errors during initial server rendering.
* Remediation agents now retry transient model stream failures while keeping quota and authentication failures terminal.
## New features
**Problem Details redesign and agent export.** Problem pages now use stable `PRB-n` references, clearer evidence sections, and bulk export actions so teams can move problem context into coding agents without losing the trail.
**Remediation safety controls.** Automatic remediation now follows severity policy gates, giving teams a clearer boundary between investigation, draft work, and changes that still need review.
## Updates
**GitHub and agent onboarding reliability.** Aient now carries GitHub installation and account-binding context more reliably through agentic setup flows, including x402-backed onboarding paths.
**Problem routing and collaboration delivery.** GitHub webhook ingestion, Slack delivery, and threaded updates now recover more consistently from duplicate events, stale bindings, and temporary provider failures.
**Telemetry ingest hardening.** Public telemetry pipelines now guard oversized payloads and route collector traffic more reliably, improving trace and log availability during incident investigation.
## Fixes
* Problem disposition actions now return honest pending states instead of false timeout failures.
* Investigated problems no longer show as actively being worked when the remediation state has moved on.
* Source map verification now retries through short storage delays and replaces stale artifact hashes.
* Remediation workspaces now authenticate Git and GitHub operations more reliably before preparing pull requests.
## New features
**Bulk source map uploads.** Source map upload APIs now batch larger sets safely, use bounded chunk sizes, and recover from invalid upload paths before they block stack trace resolution.
**Problem case documents.** Problem detail pages now show a structured case document that brings evidence and remediation context into one reviewable surface.
**Stale problem auto-resolution.** Aient can now close problems that stop reproducing after enough quiet time, so teams can distinguish natural recovery from manual fixes.
## Updates
**GitHub connection recovery.** GitHub connections recover more reliably when installation details change, keeping repository suggestions and PR permissions visible.
**Telemetry explorer polish.** The logs view now handles malformed timestamps, reduces low-value metric noise, and keeps Problems above Services so investigation starts from the highest-value surface.
**Telemetry ingest reliability.** Aient more reliably receives traces, logs, and metrics from monitored systems.
## Fixes
* Problem detection now continues correctly after quiet scan windows.
* Affected user and service counts now stay aligned with the current problem state.
* Kubernetes problems now coalesce by workload owner instead of unstable pod identity.
* Source map commit verification now retries through short upload delays and replaces stale artifact references before they block stack trace resolution.
## Updates
**Problem delivery is more resilient.** Missing Slack installs, stale channel permissions, and temporary telemetry query failures now degrade explicitly instead of interrupting problem delivery.
**Problem routing is more consistent.** Follow-up work tightened the transition between detection and routing so users see fewer stuck or duplicated problem states.
## Fixes
* Repository discovery now preserves Git credentials during fetches, reducing failed repository setup attempts.
* Slack follow-up threads now reset correctly after a constrained reply.
## New features
**Environment selector management.** Teams can manage environments from the selector, making it easier to switch between telemetry scopes without losing context.
**User account MFA management.** Account settings now include user-managed MFA controls, strengthening account security without requiring support intervention.
## Updates
**Telemetry and operations dashboards improved.** Dashboard overview latency was reduced, telemetry trace empty states were fixed, and operators have a clearer view of system health.
**Source map resolution is more accurate.** Server bundle uploads, Turbopack aliases, and chained package maps now resolve more reliably, improving the stack traces Aient uses during triage.
## Fixes
* Slack account-link recovery and PR thread replies were hardened.
* Web chat replies no longer stall after stale background work.
* Malformed authentication payloads are redacted from logs.
## Updates
**Copy your organisation ID from settings.** The organisation settings page now shows your organisation ID in a dedicated panel with a one-click copy button. Use it when filing a support request or correlating webhook payloads, where the same UUID appears as `organisationId` and in the CloudEvent `source` URL.
## New features
**Public event registry at `registry.aient.ai`.** CloudEvent `dataschema` URLs and event aliases now resolve from a dedicated public registry, independent of the app, with stable cache headers and immutable schema revisions. See the [event registry reference](/reference/event-registry).
**Aient end-to-end PR checks expose cache and timing evidence.** PR E2E runs now show cache evidence, build cache flags, and runner startup timing in the check summary so you can see what was reused between runs.
**Standalone integration docs.** Aient docs now run as a standalone Mintlify site focused on telemetry SDKs, source maps, OTLP ingest, and webhook event schemas.
## Updates
**Slack thread routing improvements.** Aient now reuses existing PR thread bindings for incoming Slack and GitHub PR events, reducing duplicate threads and dropped messages.
## Fixes
* PR E2E checks now finish cleanly when the runner disappears, instead of stalling.
* PR E2E check summaries now show an explicit "preparing the test runner" status during startup, so you see meaningful progress before the runner reports its first phase.
* Stale PR threads are no longer recreated when a PR comment arrives after the original thread bootstrap.
* Provider error qualifiers and details are preserved end-to-end in OpenTelemetry traces, making LLM provider errors easier to debug.
* Populated bars now render correctly on the problem histogram chart and sparklines after timestamp bucket alignment.