Here is what we shipped to customers during the week of May 11, 2026: 2 new knowledge sources (Linear and Notion), access-token authentication for the annie-cli so it can run headlessly from CI, and a prompt-level guardrail that bans alert silencing as a remediation. Across 7 product repos, the week landed 20 customer PRs into the Anyshift Product Updates series, picking up where last week's post left off.

New knowledge sources Annie can query

Linear

Linear joins the knowledge connectors. Add a workspace URL key and an API key in the knowledge integrations page; Annie then reaches Linear tickets, comments, and links through MCP. The UX is Jira-parity: list view with a status badge, create / edit / revoke flows, and a feature-flag gate for staged rollout. The integration ships top to bottom across 6 repos in one week: credential storage in anyshift-commons, backend CRUD and GraphQL validation in anyshift-backend, MCP wiring in ai-workbench and agent-gateway, a dedicated lambda in agent-runtimes, and the UI in anyshift-frontend.

Notion

Notion lands on the same end-to-end pattern. Drop in a Notion token from the knowledge integrations page, and Annie starts pulling runbooks, design docs, and linked pages into root-cause investigations through a Notion MCP proxied via agent-gateway to a lambda agent. The UI surface lives alongside the existing knowledge connectors, with the same CRUD lifecycle and translation strings as Linear.

Access tokens for annie-cli

The annie-cli now authenticates via long-lived access tokens, so a CI step or an Argo Workflow can call Annie without a browser session. Mint a token under the API tokens section in the web UI (project-scoped or org-scoped, read or full permissions), then pass it as a bearer credential to the CLI. Token storage uses HMAC-SHA256 with a server-side pepper, the same pattern GitHub and Stripe use, so even a database leak does not reverse the hashes. Upgrade to annie-cli v0.7.9 to unlock the headless flow.

Annie won't recommend silencing your alerts

A guardrail landed in both master investigation prompts (chat and RCA): Annie can no longer recommend muting, snoozing, or disabling monitors as a remediation. If a finding looks like a false positive, Annie now expresses it as a hypothesis with evidence rather than a "mute it" remediation. The chat prompt keeps a narrow carve-out for genuine temporary silencing after explicit user confirmation. Reliability over short-term quiet.

Also this week

  • Per-MCP error reporting: when an MCP integration fails mid-investigation, the failed_mcps log now surfaces the specific error per MCP rather than a generic failure, so customers see exactly which connector tripped and why.

FAQ

What's new at Anyshift the week of May 11, 2026?

Linear and Notion are now first-class knowledge sources for Annie. The annie-cli gained access-token authentication so it can run headlessly from CI. Annie no longer recommends muting, snoozing, or disabling monitors as a remediation.

How does the Linear integration work?

Connect a Linear workspace under the knowledge integrations page with a workspace URL key and an API key. Annie then reaches Linear tickets, comments, and links through MCP. The integration is feature-flag gated and follows the same UX as the existing Jira knowledge connector.

How does the Notion integration work?

Add a Notion token in the knowledge integrations page. Annie queries the linked Notion workspace through a Notion MCP proxied via agent-gateway to a dedicated lambda agent. Pages, runbooks, and inline links all become reachable RCA context.

How does access-token authentication work for the annie-cli?

Mint a long-lived access token in the API tokens section of the web UI, then pass it as a bearer credential to the annie-cli. Tokens are scoped (project or org) and permissioned (read or full), so a CI step or an Argo Workflow can invoke Annie headlessly. annie-cli v0.7.9 ships the client-side support.

Why did Annie stop recommending alert silencing?

Muting or snoozing a monitor hides a failure instead of resolving it. Both master investigation prompts (chat and RCA) now ban muting, snoozing, or disabling as a remediation. False-positive findings are expressed as a hypothesis with evidence instead.

That's the week. Try it and tell us what it unlocks for your team. We read every reply at app.anyshift.io.