AgentCloudThe cloud agents drive themselves

Install ​

Pick the surface that fits how you work — all three drive the same API.

# macOS / Linux · amd64 or arm64 · single static binary
curl -fsSL https://api.agentcloud.ac/install.sh | sh

export CONWAY_ENDPOINT=https://api.agentcloud.ac
export CONWAY_TOKEN=<your-token>

conway status            # verify you're connected

# conway-mcp ships with the CLI install above
claude mcp add conway \
  -e CONWAY_ENDPOINT=https://api.agentcloud.ac \
  -e CONWAY_TOKEN=<your-token> \
  -- conway-mcp

# now an agent has conway_deploy, conway_add_datastore, conway_create_repo, …

# no install — just a bearer token
export TOKEN=<your-token>
curl -H "Authorization: Bearer $TOKEN" \
  https://api.agentcloud.ac/v1/projects

No token yet? An admin issues one with conway tokens issue, or mints a scoped one for a sub-agent — see Tokens & delegation.

Ship in three commands ​

conway init blog                     # create a project (idempotent)
conway add postgres                  # DATABASE_URL injected into every service
conway deploy --dir ./blog           # build · health-gate · live URL
# → https://app-blog.apps.agentcloud.ac

Builds need no config: a Dockerfile is used if present, otherwise railpack autodetects Node, Python, Go, Rust, and more. The rollout is health-gated and auto-rolls-back if your app doesn't come up — your previous version keeps serving the whole time.

Prefer git? Push to the built-in Forge and every commit auto-deploys:

conway repos create blog --deploy-service web
git push https://<token>@git.agentcloud.ac/blog/blog.git   # → deploys on push

Everything in one platform ​

• Deploys — source → live HTTPS URL, zero config, health-gated with auto-rollback and SSE build logs.
• Datastores — managed Postgres, Redis, and Mongo; credentials generated and injected; browse tables and run queries from the dashboard.
• Git host — per-project repos where your Conway token is the git credential; git push auto-deploys, each release labelled with its commit message.
• Agent-native — one token + a fetchable skill.md is the whole onboarding; identical semantics across CLI, MCP, and REST.
• Environment & secrets — per-project variables injected into every service, versioned with each release.
• Tokens & audit — capability-scoped tokens with safe delegation, an append-only audit log, and structured {code, message, hint} errors.
• Real bare metal — Kubernetes on Talos Linux, BuildKit + railpack builds, a Caddy edge with automatic TLS. No cold starts, no black boxes.