One standard for how your engineers work with AI.
Every developer sets up Claude, Copilot, and Cursor their own way, and no one can see the sprawl. baseline turns it into a single versioned standard, ships it to every repo and laptop, and shows you exactly who's on it.
| Repo | Pack | Ver | State |
|---|---|---|---|
| acme/web | software-engineer | 1.2.0 | On baseline |
| acme/api | software-engineer | 1.2.0 | On baseline |
| acme/payments | software-engineer | 1.1.0 | Behind |
| acme/ml-platform | token-efficiency | 1.0.0 | On baseline |
| acme/infra | — | — | Not yet |
| Developer | Laptop | Synced | State |
|---|---|---|---|
| Maya Chen | MacBook Pro | 2m ago | On baseline |
| Jonas Ekwueme | ThinkPad X1 | 18m ago | On baseline |
| Priya Nair | MacBook Air | 3d ago | Behind |
| Diego Santos | MacBook Pro | 5m ago | On baseline |
| Sam Okafor | — | never | Not paired |
Twelve engineers, twelve ways of working.
AI showed up one laptop at a time. Everyone picked their own tools and wrote their own rules, so nothing matches — and none of it is visible. You can't standardize what you can't see, and you can't improve what isn't standard.
| Developer | Tool | Config | State |
|---|---|---|---|
| Maya Chen | Claude | CLAUDE.md · 340 lines | Stale |
| Jonas Ekwueme | Copilot | instructions · empty | Missing |
| Priya Nair | generic | AGENTS.md · her own | Off-book |
| Team B | Cursor | .cursorrules · from a blog | Unverified |
| Infra | Gemini | GEMINI.md · unowned | Orphaned |
Working well with AI is a standards problem, not a model problem.
The leverage was never a smarter model. It's agreeing on how your team works with AI — then keeping every repo and every laptop on that agreement as it changes. baseline is where that standard lives, ships, and holds.
Audit, standardize, distribute, govern.
Four steps, each producing something concrete the next one builds on.
Audit
Point baseline at a repo. It reads the stack and names the pack that fits.
langs: ts, tsx
tests: vitest ✓
→ software-engineer
Standardize
Pick a pack, or shape your own. One source renders config for every AI tool.
CLAUDE.md
.claude/agents/
copilot-instructions.md
GEMINI.md
Distribute
Open a pull request in every repo. The desktop app carries it to laptops, leaving local work alone.
✓ PR #128 → acme/web
✓ PR #129 → acme/api
Govern
See which repo and which laptop is on baseline, at which version, right now.
maya on baseline
acme/infra not yet
Know exactly who's on baseline.
Not a claim in a slide about "AI transformation." A live count of how much of your codebase — and how many of your engineers' laptops — run the standard you set. Down to the repo, down to the person.
| Repo / Dev | Pack | Ver | PR | State |
|---|---|---|---|---|
| acme/web | software-engineer | 1.2.0 | #128 | On baseline |
| acme/payments | software-engineer | 1.1.0 | #131 | Behind by one |
| maya · laptop | software-engineer | 1.2.0 | — | On baseline |
| priya · laptop | token-efficiency | 0.9.0 | — | Behind |
| acme/infra | — | — | — | Not proposed yet |
Your standard, versioned like code.
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. Versioned, reviewed, and rendered for every tool. Start from a curated one, drag your own together in the visual builder, or describe what you want and refine the AI's validated draft — nothing saves until you approve it.
Software-engineer harness
The disciplined default: plan, build, then an independent review. TDD, scoped sessions, and guardrails that keep changes surgical.
Token-efficiency harness
Cuts AI spend without cutting output: session hygiene, model tiering, thinking caps, and a budget on every autonomous run.
Builder / checker loop
Two roles, one rule: an independent checker re-runs the tests itself before any change counts as done.
A pack carries more than config.
Config is instructions, rendered once. A harness is state that accumulates over the life of a repo — memory, a task ledger, a wiki, a design system, a codegraph — held by the pack and kept current as things change. Turn on what you need; each capability names its own method, and one that isn't built yet says so instead of pretending.
Facts that survive the session
/remember writes a durable entry; session start reads the relevant ones back in. Org-shared memory syncs to every laptop through the agent; repo-local memory ships in the PR.
One DESIGN.md, on-brand everywhere
Theme, palette, type, components, do's and don'ts — the nine-section spec AI tools already read — so generated UI matches your brand instead of a generic template.
A ledger that outlives the run
Plans and progress recorded as they happen, so a session that gets interrupted — or a different engineer entirely — can pick up exactly where it stopped.
Read before, write after
Agents check the wiki before touching a file and update it after. Most knowledge tools only read; this one writes back, so it doesn't go stale.
A map, not a guess
A queryable map of how the codebase actually connects, so agents stop re-deriving structure they could just look up.
Your setup, on every machine you touch.
- one sourceEvery tool from one file. Claude, Cursor, Copilot, and Gemini configs render from the same pack — no more hand-syncing five files.
- the CLIRuns in your repo. baseline audit and apply show you every change before a single file moves.
- the appPairs once, stays current. A small desktop app keeps your laptop on the org standard and never overwrites the tweaks you've made yourself.
Put your whole org on baseline.
Sign up, publish your first pack, and watch it reach your repos and laptops. Free to start.
You’re on the list — we’ll reach out. Or email us.