Up and running in two minutes.

1 · Get access

Regente is rolling out through the waitlist. Join to get your invite:

The CLI and the desktop app share one engine, so humans, agents, and the dashboard read the same locked source of truth. Local-first, no account, MIT licensed.

The first time the desktop app opens, a full-screen welcome wizard walks you through setup and verifies each step live against the engine: the install step ticks only when regente install has actually landed in the repo, and the connect step only when a real agent is online. It's skippable, runs once per user (remembered in ~/.regente/onboarded), and the dashboard Overview keeps a self-verifying Setup Checklist until the workspace is wired and an agent has connected — then it disappears.

2 · Connect your agents

From the repo you want to coordinate, install the adapters once:

regente install
regente doctor --repair

regente install is end to end: it creates .regente/ if needed, writes session hooks, the MCP server, and a regente agent skill for every supported vendor (Claude Code, Codex, Gemini CLI, and Cursor), verifies each wiring step with a per-tool checkmark, and then waits until the first agent actually connects — so success means verified, not assumed (--no-wait skips the wait, --wait-minutes N changes it, --json prints machine output). Hook-wired agents auto-join the workspace when they start; Cursor gets project MCP plus onboarding rules and joins on its first Regente tool call. Already-running sessions need a restart (or a fresh terminal) before the hooks and MCP load — regente prompt prints the attach prompt to paste into any session in the meantime, and the dashboard has a copy button for it. See the exact adapter matrix →

The install step also writes Regente onboarding files for Claude, Codex, Gemini, Cursor, and the shared repo — they teach agents claim/release discipline, the auto-handoff queue, task results, usage reporting, and messaging idle holders. In the dashboard Overview, the Setup Checklist tracks the same ground: adapters wired, agents connected, first call and announcement sent.

3 · Give every agent a lane

Start your agents in the repo — nothing changes about how you run them:

cd ~/acme-app
claude          # joins as it starts
codex           # joins too
gemini          # joins when the Gemini adapter is loaded

Then give each a role; a role owns file/path lanes (globs), so Regente can warn the moment one agent reaches into another's territory:

regente role --agent claude --role frontend                 # owns public/**, src/components/**, …
regente role --agent codex  --role backend                  # server/**, src/api/**, …
regente role --agent gemini --role debugger --paths "**/*.log,scratch/**"
regente roles                                               # who holds what, and their lanes

Presets include feature, frontend, backend, debugger, tests, reviewer, docs, and infra — or pass any custom role with your own --paths. An agent holds one role at a time. More on roles & lanes →

4 · Watch collisions resolve themselves

Each agent claims a file before editing. The instant a second agent reaches for a claimed file, Regente blocks it pre-merge, queues it, and the moment the holder releases, auto-hands it to the next agent in line — with a notification, no human in the loop:

# the moment a second agent reaches for a claimed file:
⛔ blocked · pre-merge — claude vs codex on public/index.html
⤳ codex queued #1 · auto-handoff on release

# ...the instant claude releases:
↳ auto-handed off public/index.html to codex (was waiting)

See it live in the desktop command center (the fleet, collision band, and handoff queue), as a native notification even with the window closed, or right in the terminal:

regente watch        # live fleet, claims, conflicts & queue — refreshes in place
regente status       # one-shot snapshot of the same

If an agent or wrapper knows its quota, report it with regente usage; the Overview shows token and context bars per agent and leaves unknown values unguessed.

regente usage --agent codex-api --tokens-used 12000 --tokens-limit 50000 --context-used 32000 --context-limit 200000

The blocked agent isn't stuck — it gets an actionable resolution (who holds the file, its place in line, other lanes it owns) and is handed the file automatically. Participating agents coordinate before accidental same-file overlap lands.

Need finer grain? An agent can claim a single function instead of a whole file, so two agents edit two functions of the same file without colliding — and Regente warns when two agents hold different files that import each other:

regente symbols api.js                          # claimable functions/classes + line ranges
regente claim api.js --symbol login --agent alice
	regente claim api.js --symbol logout --agent bob   # same file, different function → separate symbol claims

Symbol-level claims & semantic coupling →

5 · Call agents and announce status

When one agent needs another to pick up work, use regente call. It creates an assigned task, sends a direct call message, and records a handoff so the target knows what to report back. Use regente announce for fleet-wide status changes.

regente announce "Release verification is starting" --from codex-release-lead
regente call codex-qa-verification "Verify installer" \
  --from codex-release-lead \
  --summary "Run the installer checks and review desktop launch." \
  --next "Report pass/fail in Regente."

6 · Make claims enforced (optional)

By default claims are cooperative. For hook-guarded agents (Claude Code and Gemini CLI), the live edit guard also auto-claims a file for whichever agent makes the first real edit (released when that session ends; disable with "auto_claim": false in .regente/config.json), so unclaimed collisions are caught even when an agent forgets to claim. Flip on enforcement and a pre-edit hook blocks editing a file or function another agent holds, and a git pre-commit guard blocks committing someone else's files:

regente enforce strict      # off | warn | strict

It's a guardrail for a cooperating fleet, not a security boundary. Enforcement & identity →

Verify everything's wired

regente doctor --adapters   # exact hook + MCP + guard health per tool
regente doctor --repair     # repair stale/missing repo-local adapter wiring
regente discover            # running sessions: connected / detected / external
regente metrics             # activation, cross-tool depth, collisions caught

connected means the running session has a Regente heartbeat; detected means it is running in this repo but still needs the adapter loaded or the attach prompt pasted; external means it is running in another repo.