Docs

Quickstart

Run one command in a repository your team works on:

npx @align-agent/cli init

It signs you in through your browser, shows exactly what is captured and what never is, asks you to opt this repository in, adds its hooks to Claude Code, Cursor and Codex (merged, never overwritten), and, if you agree, reads your recent sessions to find the decisions already in them. Then confirm the ones that are right:

align confirm

From the next session on, your agent starts with the confirmed decisions that apply to the files it touches. Until you confirm one, there is nothing to give it.

The docs

• Install and set up: init step by step, Claude Code, Cursor, Codex and other MCP agents
• Proposals and confirming: where decisions come from, and every way to confirm one
• Packs and MCP tools: what your agent is given, and what it can ask
• Remote MCP and OAuth: connect an agent that runs elsewhere
• GitHub: the App, verified repositories, collaborators and auto-join
• The PR check: one comment, one check, and its commands
• Slack and Linear: questions, records and briefs (add-ons, coming soon)
• Agency share links: read-only links for your clients
• Plans and billing: plans, contributors, the trial
• What is captured, and who sees it: redaction on your laptop, retention, visibility
• Deleting and exporting: sessions, accounts, organisations, exports
• CLI reference: every command and option
• Troubleshooting: the doctor's checks explained

Questions

Does ALIGN confirm decisions on its own? No. Models propose; a person confirms every decision, in the terminal, the console, a pull request or, with the Slack add-on, Slack.

Can my teammates read my sessions? No. Your raw transcripts are visible only to you. Teammates see a decision's short supporting quotes once it is confirmed.

What leaves my laptop? Everything that could leave the machine passes the redaction engine first: transcripts, commit messages, and every argument of align decide, align ask and the MCP tools. File contents never leave, including anything the agent read or wrote through tools: transcripts carry the user's and the agent's words only, fenced code blocks become ‹code: N lines›, and every tool call becomes ‹tool: <name>, <bytes> bytes› plus the repository-relative paths it touched. Tool outputs, diffs from your laptop and the shell commands tools run are never captured. Text you paste without a code fence is sent after secret redaction. The server redacts everything again, including GitHub, Slack and Linear text, before it stores it or sends it to a model. (details)

Does it slow my agent down? The hooks run in the agent's path and are kept under 200 ms at p95 (session start, which fetches the pack, under 1.5 s); align doctor --bench measures it on your machine.

How do I stop it? align off stops capture in a repository at once; align uninstall removes it.

See also Install, Pricing and Security and privacy.