@0dai-dev/cli

0dai

One config layer for six terminal AI agent CLIs — Claude Code, Codex, OpenCode, Gemini, Aider, Qoder — plus Cursor's MCP/editor bridge.

Install

npm install -g @0dai-dev/cli

Five commands that matter

cd your-project
0dai init                    # create ai/ layer, auth, and MCP bootstrap
0dai status                  # maturity, swarm tasks, and session state
0dai sync                    # refresh ai/ after repo or server changes
0dai remote push             # upload this bound project's ai/ to remote project MCP
0dai run "add a test" --now  # run one task locally now, get a scored receipt — no account
0dai doctor                  # health check for credentials, drift, and env

Run a task, open the PR

0dai run "<goal>" --now --pr runs the task on this repo and opens a pull request whose body is the scored receipt — goal, agent, per-task result, score, cost, the files that changed, and the pre-run commit to revert with.

0dai run "add password reset tests" --now --pr

The PR body reads as a review-ready summary, not a wall of logs. Labels stay honest: the status says run ✓, not tests ✓ — the score is agent task outcomes plus change detection, never a claim that a test suite went green. --pr is opt-in and requires --now; on a no-op run, or when gh is missing or unauthenticated, it prints the manual git/gh commands instead of failing.

No account is needed to run locally and read the receipt. 0dai undo reverses the run.

All commands

Start (first 5 minutes):

0dai init                           # create ai/ layer + MCP [--local] [--dry-run] [--minimal]
0dai init --auth-code <code> --code <TEAM-CODE>   # non-interactive auth + activation
0dai doctor                         # check health, credentials, and drift [--drift]
0dai doctor --security              # data-boundary disclosure (egress, local sinks, MCP, authZ)
0dai doctor --security --json      # same as a stable artifact: attach to a vendor questionnaire or diff in CI to catch new egress (host/time/path values isolated under "meta")
0dai status                         # maturity, swarm, and session state [--json]
0dai quickstart                     # auth, init, doctor, and status in one pass
0dai detect                         # show detected stack
0dai auth login                     # OAuth/device flow [--device] [--no-browser] [--code CODE] [--mcp]
0dai auth mcp                       # store MCP token from current auth or device code
0dai auth status                    # account and usage
0dai activate free                  # claim free activation license

Daily (regular work):

0dai sync                           # update ai/ after repo or server changes [--dry-run] [--yes] [--quiet] [--force]
0dai eject                          # remove 0dai from this repo and print what it touched (leaves CLAUDE.md/AGENTS.md) [--dry-run] [--yes]
0dai run <goal>                     # run one task locally with --now (scored receipt, no account); --share for a public link, --pr to open a PR [--dry-run] [--dry-cost] [--agent claude|codex|gemini]
0dai swarm status                   # show queued, active, and done tasks
0dai swarm add                      # queue one task [--task '...' --to agent]
0dai swarm pick                     # pick one queued task [--agent codex]
0dai swarm-run                      # repo-checkout helper: add, dispatch, and wait for one swarm task as JSON
0dai harvest                        # convert experience events into candidate lessons
0dai watch                          # live task monitor [--interval N]
0dai reflect                        # session reflection: delivered, delegation, blockers
0dai standup                        # repo-checkout helper: morning voice briefing about overnight agent work
0dai feedback push --target .       # send feedback to 0dai
0dai feedback retry                 # retry queued feedback after a failed push

On npm, swarm gives you the task queue plus cost telemetry. Full swarm orchestration (pick/run/quality) runs from a 0dai repo checkout with the Python toolchain — 0dai doctor lists what's missing.

Pro / advanced:

0dai init-existing                  # existing-repo setup alias for init
0dai project bind                   # bind repository to your 0dai account [--name NAME] [--json]
0dai project status                 # local project binding and health [--json]
0dai graph push                     # upload local graph (Pro: edges, Free: nodes)
0dai graph pull                     # download server graph and merge locally
0dai graph status                   # local graph stats and sync state
0dai heatmap                        # repo treemap: LOC × agent-edit intensity
0dai session save                   # save session for roaming
0dai provider                       # provider profiles and local BYOK registry [status|list|switch|clear]
0dai models                         # model ratings (--fast/--balanced/--deep/--available)
0dai quota                          # agent subscription usage [--refresh] [--json]
0dai usage                          # local token, task, and USD ledger [status|daily|monthly]
0dai workspace                      # tmux workspace sessions (init|up|status)
0dai report                         # repo-checkout helper: privacy-safe project reports (preview|push|status)
0dai audit-log verify               # verify ai/manifest/audit.jsonl hash-chain integrity [--json]
0dai compliance                     # SOC2/ISO evidence and ADR audit-trail export
0dai experience                     # structured experience events (list|stats|record-json|sync|warnings|dismiss)
0dai receipt                        # session receipt PNG [--last|--active|--session ID]
0dai peer ls                        # exact broker-visible sessions
0dai peer inbox --unread            # metadata-only private inbox listing
0dai peer send --to <session> --priority info --text <text> [--wait ack --timeout 30s]
0dai peer read <message-id>         # fetch body and record observed
0dai peer ack <message-id>          # record an explicit acknowledgement
0dai peer reply <message-id> --text <text>
0dai gh branch-protection           # GitHub branch protection [print|apply|install]
0dai import claude-code-agents      # import .claude/agents/*.md as personas [--dry-run]
0dai auth logout                    # remove credentials
0dai activate code <TEAM-CODE>      # redeem Pro/Team activation code
0dai activate status                # activation and bound-project status
0dai redeem <CODE>                  # redeem a plan upgrade code
0dai upgrade                        # open pricing page (browser or printed URL)

Global flags: --target PATH, --version, --help, --json, --quiet. See 0dai --help for the full surface.

0dai peer is the canonical Session Bus namespace. 0dai session send ... is an exact compatibility alias to peer send; all other session commands remain roaming commands. Peer messages go through the Runtime Broker and a private Unix socket bound by root-owned host policy; project configuration, PATH, and ODAI_BROKER_SOCKET cannot redirect it. The npm package neither downloads nor starts that daemon automatically. Install the verified native companions and provision the broker UID, host/authorization policy, private state, exact registration, and both filesystem plus owner-side tmux server-access permission for the exact managed socket explicitly; see docs/migration/go-packaging.md. If the broker is unavailable, the command fails without claiming a receipt or falling back to direct tmux input.

Governance & security

These ship today. They are not on the short --help (which keeps to the five core commands) — find them under 0dai --help --all. Built for the people who have to answer a security questionnaire or sign off on what a tool talks to.

0dai trust --json                   # pre-run blast radius: protected paths, agent authority, declared egress
0dai doctor --security --json       # data-boundary disclosure: egress endpoints, local sinks, MCP, authZ tiers (run after 0dai init)
0dai compliance --framework soc2 --json     # SOC 2 evidence: audit log, access control, licensing, ADR trail
0dai compliance --framework iso27001 --json # same evidence mapped to ISO 27001:2022

• 0dai trust — what a run can touch before it runs: the protected-path patterns, the agent authority matrix, and the exact hosts the CLI is allowed to reach. --json for a diffable artifact.
• 0dai doctor --security — the data-boundary disclosure: which endpoints the CLI calls, where it writes locally, which MCP servers are wired, and the authZ tiers (default-deny). The JSON form is stable — attach it to a vendor questionnaire, or diff it in CI to catch a new egress endpoint. Run it inside a repo that has had 0dai init.
• 0dai compliance — SOC 2 or ISO 27001 evidence pulled from the local audit log, write-ahead log, role policy, and ADR trail. --json to export.
• Egress allowlist — ai/manifest/egress-allowlist.json lists the hosts the CLI may talk to (api.0dai.dev, mcp.0dai.dev by default). 0dai doctor --security fails if a disclosed endpoint resolves to a host not on the list. Self-hosters set ODAI_API_URL / ODAI_MCP_HOST and list their own hosts. Delete the file to opt out.
• 0dai sync --check / --install-hook — --check is a read-only CI gate that exits non-zero when the committed ai/ layer has drifted from the manifest. --install-hook wires the same check into a git pre-commit hook so drift never lands.
• 0dai eject — remove the 0dai machinery and print exactly what it touched (reverses what init/sync wrote; leaves CLAUDE.md/AGENTS.md). --dry-run previews the clean exit.

Power commands

0dai swarm quality                  # quality breakdown across completed swarm tasks
0dai swarm sessions                 # live session registry: idle/busy/stale per session [--json]
0dai receipt                        # render a session receipt PNG
0dai receipt share latest           # publish the last scored run receipt to a shareable link (asks first)

• 0dai swarm quality — score and outcome breakdown across the tasks the swarm has finished, so you can see where agent work actually held up.
• 0dai swarm sessions — the live session registry as a table: which lanes are idle, busy, or stale. --json for the raw view.
• 0dai receipt — render a session as a 1200×630 PNG. receipt --run renders the scored run receipt instead; receipt share <run_id|latest> publishes a scored run to a public link, and asks before it posts anything.

Optional: terminal sessions

0dai terminal launch --tool codex runs another CLI inside a real pseudoterminal you can attach to and detach from. It needs node-pty, an optional native dependency that sometimes fails to compile on npm install — nothing else in the CLI depends on it. If it's missing, 0dai terminal launch tells you and exits; 0dai doctor also flags it under "terminal sessions" with the exact fix for your platform, roughly:

sudo apt-get install -y build-essential python3   # Linux
xcode-select --install                            # macOS
# Windows: Visual Studio Build Tools, "Desktop development with C++"
npm install -g @0dai-dev/cli

What it does

0dai init and 0dai sync are activation-first. init can complete auth and plan activation in one run: browser/OAuth exchange code via --auth-code, Team/Pro activation code via --code or --activation-code, or the interactive OAuth/device-code prompt. After that it binds the project and sends only allowlisted project metadata (file names + hashes of package/build manifests — never their contents) to the API and generates:

• ai/ — manifests, personas, skills, playbooks, delegation policy
• .claude/ — settings, agents, hooks, rules
• .codex/ — config, agents
• .gemini/ — settings, agents
• .aider/ — config, agents
• AGENTS.md, .mcp.json

Your source code is never sent. Only file names and hashes of package/build manifests — never their contents.

Privacy

Local activation and experience records stay on your machine: ai/meta/telemetry/activation.jsonl, ai/experience/, and the generated ai/ layer/project agent configs are local files. Server-side activation sends one free_tier_activated event when you run 0dai activate free unless telemetry is opted out. See https://0dai.dev/legal/privacy.

Why 0dai, not just Cursor or Copilot?

Cursor and Copilot are editors. They help inside one coding session. 0dai writes a project layer that Claude Code, Codex, OpenCode, Gemini, and Aider can all read. The point is not another autocomplete box; it is one manifest, one set of agent roles, one task queue, and one health check for the repo.

A fresh 0dai init turns a repo into an AI-ready workspace. It records stack detection, safe project commands, agent preferences, MCP config, and delegation policy under ai/, then writes native files such as AGENTS.md and .mcp.json. Your source stays local; only file names and hashes of package/build manifests go to the API — never their contents.

0dai is useful when work needs more than one model or more than one sitting. You can ask for a backlog item with 0dai run "add password reset tests", check the queue with 0dai swarm status, and continue from the same project facts in another agent CLI. Cursor or Copilot can still be your editor; 0dai keeps the repo-level context and task state outside the editor.

The first five minutes are direct: run 0dai init, run 0dai doctor, then try one backlog task with 0dai run "..." or inspect state with 0dai status. If you only want a local setup, use 0dai init --local; cloud mode adds sync, graph history, and shared task records.

Free-tier examples

On the free tier you get 1 swarm task/day — an activation teaser (Pro lifts this to 50/day). Enough for one daily cleanup pass; pick a task to queue, for example:

• Write missing tests for the login form.
• Add empty-state copy to the dashboard.
• Check a failing CI log and suggest the smallest code fix.
• Draft a PR checklist from the files changed.
• Summarize today's ai/swarm/done tasks for standup.

Each backlog item splits into one or more agent prompts. The free tier's 1 task/day suits a daily cleanup pass; Pro (50/day) is for larger task queues, graph sync with edges, and session roaming across projects.

Links

• Website: https://0dai.dev
• Docs: https://0dai.dev/docs
• API: https://api.0dai.dev