obi

Command-line client for the Obitrain API — one generic, scriptable interface to every endpoint, for interactive use and automation.

Instead of wrapping each endpoint in its own subcommand, obi gives you three primitives:

• obi auth — authenticate once, stay authenticated
• obi api — call any endpoint (method, query, body, headers)
• obi schema — discover endpoints, parameters and payload shapes offline

Install

uv tool install obitrain      # exposes the `obi` binary
uvx --from obitrain obi --help

User quickstart

obi auth login

obi auth login prints a short code (like QX2F-LX4T). Enter it in the Obitrain mobile app under Account → Link a device, and the CLI receives its own API token — no password typed in the terminal, and it works for Google/Apple accounts too. The token is long-lived and revocable from the app at any time.

Then call anything:

obi api /v1/user                                  # your profile
obi api /v1/training/sessions -q limit=5          # last sessions
obi api /v1/user -X PATCH -d '{"lang": "fr"}'     # update a field
obi api /v1/training/session -d @session.json     # body from a file (implies POST)

Continue with the user quickstart or see the full documentation.

Authentication

Command	Purpose
obi auth login	Device-code login: approve in the app, token stored in the active profile
obi auth set <token>	Store a token generated from your Account page
obi auth status / whoami	Inspect the active profile / authenticated user
obi auth token	Print the token, for scripting
obi auth clear	Remove stored credentials

• Profiles — --profile work (or OBI_PROFILE) keeps separate accounts; credentials are stored per profile with 0600 permissions. obi auth profiles lists them.
• Ephemeral tokens — OBI_TOKEN / --token authenticate a single invocation without writing anything to disk; ideal for CI.
• Base URL — defaults to https://api.obitrain.com; override with --base-url or OBI_BASE_URL.

Calling the API

obi api <path> [-X METHOD] [-q k=v ...] [-d BODY|@file|@-] [-H k:v ...] [-o json|pretty|raw|yaml] [-n]

• -d implies POST unless -X says otherwise; @file and @- (stdin) are supported.
• -n / --dry-run prints the fully resolved request (token redacted) without sending it.
• Response bodies always go to stdout; diagnostics go to stderr as one-line JSON.
• Output is human-friendly by default (syntax-highlighted on a TTY) and plain JSON whenever output is piped or an agent environment is detected; --json (or -o json|raw|yaml) forces a format explicitly.

Exit codes

Code	Meaning
0	Success — parse stdout
1	Usage error
4	Authentication required or failed
5	Network error (connection, DNS, timeout)
6	Server error (5xx)
7	Client error (4xx)
130	Cancelled (Ctrl-C)

Discovering the API

The OpenAPI spec is bundled with the package — discovery works offline:

obi schema tags                              # tags with operation counts
obi schema list --grep session               # find operations
obi schema show /v1/training/sessions -X GET # params, body & response schemas, inlined models
obi schema show /v1/stats/activity/weekly    # concrete paths resolve to their {template}

Agent quickstart

obi was also designed for reliable use by AI agents. The machine-facing contract includes:

• Self-documenting binary — obi quickstart prints the full agent guide (discovery → auth → calls → error repair) to stdout; obi --help-json gives the machine-readable command tree. No repo or web access needed.

• Deterministic output — bodies on stdout, one-line JSON diagnostics on stderr, exit codes above. Pass --json to force plain JSON; in practice you already get it, since output falls back to JSON when an agent environment is detected (CLAUDECODE, CURSOR_AGENT, GITHUB_COPILOT, AMAZON_Q, OBI_AGENT_MODE, …), when NO_COLOR is set, or when stdout is not a TTY.

• Errors carry their own fix — 404/405/422 diagnostics include a hint field built from the bundled spec:

  $ obi api /v1/stats/activity/weekly
  {"error": "http_error", "status": 422, "method": "GET", "path": "/v1/stats/activity/weekly",
   "hint": "required params: range_type (path, enum: daily|weekly|monthly), from_date (query, date),
   to_date (query, date); see `obi schema show '/v1/stats/activity/{range_type}'`"}
  
  $ obi api /v1/stats/activity/weekly -q from_date=2026-01-01 -q to_date=2026-06-01   # repaired

The recommended loop: obi schema list --grep … → obi schema show <path> → obi api … -n (optional dry-run) → obi api … → on exit 7, read hint and retry.

Works with coding agents. Any agent that can run a shell can drive obi — give it a token via OBI_TOKEN and tell it to run obi quickstart. Claude Code is auto-detected (CLAUDECODE); for OpenAI Codex (AGENTS.md) and Mistral Vibe, set OBI_AGENT_MODE=1 to force JSON:

OBI_TOKEN="<api token>" claude                                   # auto-detected
OBI_AGENT_MODE=1 OBI_TOKEN="<api token>" codex exec "use obi to fetch /v1/user"
OBI_AGENT_MODE=1 OBI_TOKEN="<api token>" vibe -p "use obi to fetch /v1/user" --enabled-tools "bash*"

See the agent quickstart for the complete workflow, including per-tool setup.

Development

uv sync --all-groups
uv run ruff check src tests && uv run ruff format --check src tests
uv run pyright
uv run pytest

Generated models

src/obitrain/api/models.py holds TypedDicts generated from static/openapi.json (the vendored OpenAPI 3.1 spec). Regenerate after updating the spec:

sh bin/codegen.sh
git diff --exit-code src/obitrain/api/models.py   # CI fails if these drift

To refresh the spec from upstream, replace static/openapi.json and rerun codegen.

Docs

uv run --group docs zensical serve   # live preview
uv run --group docs zensical build   # static site -> ./site

Full docs: https://obitrain.github.io/obi/.