You are an AI agent operating mago on behalf of a human (the CEO). Drive everything via the
mago and gh CLIs. The human only: pays, installs the GitHub App, answers HITL/clarify questions, and says "go".
gh auth setup-git once so the worker can push
branches to private repos (public repos work without it).OPENCODE_API_KEY
(or your provider's key), then set provider: pi and model: opencode-go/deepseek-v4-flash
(or any provider/model) in each agent file.claude on PATH and logged in (your Claude subscription),
then run with MAGO_PROVIDER=claude MAGO_MODEL=sonnet. No API key.curl -fsSL https://mago.intrane.fr/install.sh | sh
mago register --email you@co.com --password <pw> # account + 48h NO-CARD trial (license cached) mago account status # -> plan: trial (active, ~Xh left), license_key mago subscribe # after the trial: Stripe link, the HUMAN pays (€20/mo) mago billing # manage/cancel the subscription (Stripe portal)
The 48-hour trial runs the full loop immediately — no card. Subscribe anytime to continue past it.
The human installs the mago GitHub App on their repos (one browser click). Then:
mago link --installation <id> # entitles your repos (id is in the install URL) mago link list # confirm entitled repos
mago init ./company mago project add <name> --repo owner/repo # the repo your agents work on AND watch for issues MAGO_TASK_LABEL=mago mago serve --relay -C ./company # only act on issues labeled "mago"
Single repo (default): a lone project repo becomes your backlog automatically — its
issues are the work, PRs close them directly. Multi-project (one worker, several repos): set
MAGO_GH_REPO to the backlog/command repo, mago project add the others, then file
issues there labeled project:<name> (mago task add "…" --project <name>) to
open the PR on that project's repo. Note: proactive planning fills the backlog repo only — multi-project is for reactive dispatch.
MAGO_TASK_LABEL=mago the worker only takes issues labeled mago — so you can point it at a real repo safely; labeling an existing issue picks it up.mago:clarify → the planner posts a plan + open questions; answer in comments (N rounds); add mago:go to implement.mago:in-progress, mago:blocked, mago:hitl (needs the CEO), agent:<name>, project:<name>.← mago.intrane.fr · machine-readable version: /llms.txt