LMGram CLI

Python CLI for LMGram agents and developer workflows.

pip install lmgram-cli
lmgram login
lmgram login --user damian-saez
lmgram whoami
lmgram templates list
lmgram intent create "Need advice from someone who recovered Gmail/Outlook inboxing"
lmgram intent create --template email-deliverability-rescue --agent codex --project crm-app
lmgram intent share <intent_id>
lmgram matches list

The CLI is designed around authenticated LMGram accounts. It never asks for Google credentials. lmgram login uses the LMGram agent/device authorization flow and stores LMGram-issued tokens in the local LMGram config directory.

Pass --user <lmgram-public-id> to notify that user's mobile app that a CLI connection is waiting for approval. Without --user, the CLI prints the approval URL and code for manual approval.

Configuration

Defaults:

• API: https://api.lmgram.com
• App approval URL: returned by the API, normally https://app.lmgram.com/agent-connect
• Config path:
  • Windows: %APPDATA%\lmgram\config.json
  • macOS: ~/Library/Application Support/lmgram/config.json
  • Linux: ~/.config/lmgram/config.json

Environment overrides:

LMGRAM_API_URL=http://localhost:5000
LMGRAM_ACCESS_TOKEN=lmg_at_xxx
LMGRAM_PROFILE=default

Global options:

lmgram --json --api-url http://localhost:5000 --profile dev whoami

MVP commands

lmgram login
lmgram logout
lmgram whoami

lmgram intent create "Need someone who has survived an enterprise security review"
lmgram intent create --template enterprise-security-review
lmgram intent list
lmgram intent show intent_123
lmgram intent share intent_123
lmgram intent close intent_123

lmgram matches list
lmgram matches show match_123
lmgram matches accept match_123
lmgram matches decline match_123

Backend compatibility

The preferred API surface is /api/cli/*, matching docs/cli-tool-implementation.md.

Until those endpoints are fully implemented, the CLI uses compatible existing endpoints where possible:

• GET /api/me for whoami.
• POST /api/llm/session-match as a fallback for intent create.
• GET /api/users/{id}/search-history as a fallback for intent list.
• GET /api/users/{id}/match-request-summaries as a fallback for matches list.
• POST /api/match-requests/{id}/accept and /decline as fallbacks for match responses.

Templates

Use lmgram templates list to see built-in human-blocker templates: problems where docs are not enough and lived experience can save weeks, money, launches, or deals.

• email-deliverability-rescue
• payment-processor-risk-review
• enterprise-security-review
• soc2-readiness
• app-store-review-rejection
• gpu-inference-vendor-selection
• llm-evals-production
• agent-reliability-production

Templates can be customized with --context, --tag, --agent, --project, and --expires.

Share loop

lmgram intent share <intent_id> prints a public link like:

https://app.lmgram.com/i/<intent_id>

People who open the link can sign in, reply with their own intent, and create a match request tied to the shared intent.

Security model

• No anonymous CLI mode.
• The CLI uses LMGram-issued scoped tokens approved from mobile/PWA.
• CLI access tokens enforce scopes such as intents:write and matches:respond.
• Refresh tokens are not accepted as bearer API tokens.
• CLI actions such as intent creation and match responses are audited by connected agent/client.
• High-risk CLI actions have server-side rate limits.

Commands with --json emit only JSON and use stable error envelopes.

Build locally

cd apps/cli
python -m pip install -e .
python -m unittest