The full loop, in detail.
baselane is four verbs that feed each other: audit what exists, standardize it into a pack, distribute the pack everywhere, govern what happened. The engine that runs the first three is open source (Apache-2.0); the org control plane is the product. Here is each one, shown with the real surfaces.
Start from what's really there.
Point baselane at any repo. It reads the stack (languages, package manager, frameworks, test setup, mono or single) and names the pack that fits. No forms, no guesswork, nothing uploaded: analysis runs against the repo and only the profile persists.
- Never used for training. Your source is profiled and discarded.
- Fit-gated. A pack is only recommended when the repo can actually run it.
One pack. Every tool speaks it.
A pack is your team's way of working with AI, written down: the agents it runs, the commands it exposes, the rules it holds people to. Author it once. baselane renders native config for every tool your engineers actually use.
- One source of truth. AGENTS.md is canonical; everything else derives.
- Versioned like code. Packs ship as v1.2.0.
- Built visually or described in English. Drag blocks together in the visual builder, or describe the pack you want and refine the AI's validated draft. Nothing saves until you approve it.
Through the front door: pull requests.
The standard shows up as a reviewable pull request in every repo, opened by baselane. Nothing lands silently, nothing overwrites local work. Merge it like any other change; decline it and the coverage board says so.
- A PR per repo. Reviewed and merged by the owning team.
- Never clobbers. Existing files are respected, and conflicts are surfaced for you to resolve.
- Merges, never overwrites. An existing CLAUDE.md or AGENTS.md is read first; baselane's additions slot into managed regions, your own text stays exactly where you put it.
Coverage you can point at.
The dashboard shows exactly how AI adoption is going: which repo runs which pack at which version, which laptops are paired, and what's behind, down to the person.
- Repos and people. Both surfaces, one board.
- Always current. States come from merged PRs and agent syncs, never from a survey.
| Repo / Dev | Pack | Ver | State |
|---|---|---|---|
| acme/web | software-engineer | 1.2.0 | On baselane |
| acme/payments | software-engineer | 1.1.0 | Behind |
| maya · laptop | software-engineer | 1.2.0 | On baselane |
| acme/infra | — | — | Not yet |
A pack carries state, not just instructions.
Config renders once. A harness accumulates: memory that outlives a session, a task ledger that survives an interrupted run, a wiki that's read before a change and written after, a design system every tool renders against. Each capability picks its own method (file-based today, a specialized tool where one's built), and an unimplemented one says so; it never quietly does nothing.
- Memory — live. /remember writes it, session start reads it back. Org-shared memory syncs to every laptop.
- Design system — live. A nine-section DESIGN.md so generated UI matches your brand.
- Tasks and wiki — rolling out. A progress ledger and a read/write knowledge base, landing capability by capability.
- Codegraph — coming soon. A queryable map of the codebase, not a guess.
Repos are half the story. Laptops are the other half.
Machine-level setup (personal agents, commands, skills) never lives in a repo. The baselane desktop app, coming soon, will pair a laptop to your org once and keep it on the published standard automatically. Local tweaks will never be overwritten, and drift will just show up on the dashboard. Today, the open-source CLI covers this: npm install -g baselane.
See the loop run on your own repos.
Sign up, publish your first pack, and watch coverage climb. Free to start.
You’re on the list. We’ll reach out. Or email us.