Introduction
The admin-side companion to the operator introduction. This covers standing up and running a Tembo Agent Studio (TAS) instance, rather than building and running agents in one.
TAS is self-hostable first: identity, data, and runtime stay inside your environment. You deploy it on your own infrastructure, point it at a Git repo you own, set provider keys, and invite your team.
How it fits with Tembo
Section titled “How it fits with Tembo”TAS is the control plane. The authoring step — turning a chat message into a diff — is delegated to the Tembo Coding Agent Platform. You plug a Tembo API key into workspace settings, and TAS uses it to open pull requests against your repo: new agents from chat, and “improve the agent” submissions from any run. TAS calls out to Tembo coding agents the way a CI system calls out to compilers.
Until a Tembo API key is set, chat-to-PR authoring stays hidden — running existing agents only needs an LLM provider key (Anthropic or OpenAI).
Architecture in one breath
Section titled “Architecture in one breath”A Next.js web control plane, a Rust (axum + sqlx) API that owns the runtime and
Postgres migrations, and a Postgres database. Agent definitions live in your
connected GitHub repo; the API runs agents as subprocess calls into the bundled
pydantic-ai wrapper.
Ways to run it
Section titled “Ways to run it”- Local / from source —
docker compose up --buildbrings up web, API, and Postgres. Good for evaluation and development. - From prebuilt images —
compose.release.yamlpulls pinned images from GHCR instead of compiling from source. - Managed hosts — the full stack on Railway or AWS (ECS Fargate + RDS), or the web tier on Vercel with the API on a long-lived host.
What admins manage
Section titled “What admins manage”- Settings — repository, LLM provider keys, the Tembo API key, Composio, Slack apps, and branding.
- Audit & roles — the append-only audit timeline and the viewer / operator / workspace-admin model (plus instance admins).
- Slack apps — installing and scoping the workspace Slack apps that launch agents.
Deploy guides
Section titled “Deploy guides”- Setup checklist — start here. A zero-to-running checklist from accounts and keys through first sign-in.
- Deploy on Railway — full stack on Railway from published images.
- Deploy on AWS — ECS Fargate + RDS.
- Deploy on Vercel — web on Vercel, API on a long-lived host, managed Postgres.
Versioning
Section titled “Versioning”Production instances pin to explicit CalVer release tags (e.g. v2026.6.12);
released container images are published to GHCR with signatures and SBOMs. See
the Changelog for what’s in each release.