Quickstart: five minutes to a connected agent
1. Create a workspace
Sign up, name your organization, and optionally invite teammates by email. One step, under a minute.
2. Import the rules you already have
Paste or upload the agent rules your team already wrote: CLAUDE.md, AGENTS.md, Cursor rules, or similar files. Firmament splits them into individual candidate rules and shows you a review checklist: every rule is editable, scoped, and individually selectable before anything is stored. Nothing is imported without your review.
Import the instructions humans wrote for agents. Don't import facts an agent can derive from reading the code; those go stale.
Never commit directly to main; every change goes through a PR.
Company ▾Deploys need a green CI run first.
Company ▾I prefer tabs over spaces.
Just you ▾
stylized · 3 candidate rules found, 2 selected
Import 2 rules3. Connect your first agent
Your workspace's MCP page shows your MCP URL. Connect any MCP-capable agent; authentication uses your normal workspace login (OAuth).
For Claude Code:
claude mcp add --transport http firmament <your-workspace-mcp-url>
For other clients (Cursor, Windsurf, custom agents) see the MCP guide.
4. Work normally
Your agent now has two extra tools, ask and submit. Before tasks it pulls your organization's proven guidance; after tasks it reports outcomes. The learning happens server-side, in the background. Here is what a working session looks like:
5. Review what your team is learning
Lessons proposed for a team land in the team's approval queue. Managers approve, reject, or enable auto-accept per team. Approved lessons start serving every agent on the team immediately. The platform shows knowledge in three views (Personal, Team, Company) with version history and usage:
- New
Retry flaky S3 uploads with jittered backoff
via Cursor · backend
- Update
Gate deploys on make migrate-check (was: run migrations manually)
via Claude Code · platform
stylized · the real queue lives in your dashboard