← Back to research
·7 min read·opensource

Tembo Agent Studio

Tembo Agent Studio is an MIT-licensed, self-hosted control plane for team AI agents — agent definitions live in Git, every change becomes a PR, and runs, audit logs, and identity stay in your environment.

Key takeaways

  • GitOps for agents — agent definitions are files in a repo, and every change (from engineers, PMs, or end users) flows through pull request review
  • Fully self-hosted under an MIT license: runs, audit logs, and identity never leave your environment, with Docker Compose, Railway, AWS, and Vercel deployment paths
  • Very early project — open-sourced May 2026 with 3 GitHub stars as of June 2026, shipping calendar-versioned releases at a fast clip (15 releases in its first month)

FAQ

What is Tembo Agent Studio?

A self-hosted control plane for AI agents where agent definitions live in Git, changes are reviewed as pull requests, and execution, audit logs, and identity stay inside your own infrastructure.

How much does Tembo Agent Studio cost?

It is free and MIT-licensed; you pay only for your own infrastructure and model API usage. A managed or commercial tier is not publicly disclosed.

What models and surfaces does it support?

Anthropic and OpenAI model providers; surfaces include a web interface, an API, external event triggers, and per-user connections to Slack, Gmail, Google Sheets, Notion, and GitHub.

How is Tembo Agent Studio different from Tembo's coding agent platform?

Tembo's main platform orchestrates coding agents that ship PRs; Agent Studio is the open-source control plane for governing general-purpose team agents — definitions, runs, audit, and identity — on your own infrastructure.

Executive Summary

Tembo Agent Studio is a self-hosted control plane for AI agents, built by Tembo — the company behind the coding agent orchestration platform of the same name.[1] The pitch is GitOps for agents: agent definitions are files in your repo, every change becomes a pull request, and runs, audit logs, and identity stay in your environment. Where most team agent platforms are hosted SaaS with prompts edited in a web form, Agent Studio treats agents like production software — versioned in Git, reviewed before deploy, and auditable after the fact.

It is also very new. The repository was created in May 2026, carries an MIT license, and has 3 GitHub stars as of June 2026 — but it shipped 15 calendar-versioned releases in its first month, with the last push two days before this writing.[2][1]

AttributeValue
CompanyTembo
ReleasedMay 2026
GitHub Stars3 (as of June 2026)
LicenseMIT
Primary LanguageTypeScript (Next.js frontend, Rust API backend)

Product Overview

Agent Studio is the layer between "someone wrote a prompt" and "the team runs a fleet of agents." Agents are defined as files in a Git repository; the studio renders them into a shared web workspace where teammates can run them on demand, on schedules, or from external event triggers.[1] All modifications — whether from engineers, PMs, or end users giving feedback on a run — flow through pull request review, so agents can evolve while governance stays explainable.[3]

The Tembo platform itself closes the authoring loop: plain-language requests are converted into reviewable pull requests against the agent repo, so non-developers can propose agent changes without touching code directly.[3]

Key Capabilities

CapabilityDescription
Git-based definitionsAgents are files in your repo; every change is a PR
Runs & automationsOn-demand runs, schedules, and external event triggers
Per-user connectionsSlack, Gmail, Google Sheets, Notion, and GitHub connections scoped to each user
Feedback loopsUsers open PRs from run feedback to iterate on agents
Identity & RBACSign-in via Google, Microsoft Entra ID, or generic OIDC; audit logs and role-based access
Python toolsCustom Python tools attached to agents

Product Surfaces

SurfaceDescriptionAvailability
Web workspaceShared UI for running agents and reviewing runsAvailable
APIProgrammatic runs and event triggersAvailable
SlackPer-user connection for agents that read/act in SlackAvailable
DiscordNot supported

Technical Architecture

Agent Studio is a self-hosted TypeScript application (85.7% of the codebase) with a Next.js frontend and a Rust API backend, plus Python for agent tooling.[1] Deployment is Docker Compose — from source or prebuilt GHCR images — with documented guides for Railway, AWS, and Vercel. Quick start: configure .env with secrets, enable at least one sign-in provider, docker compose up.

Key Technical Details

AspectDetails
DeploymentSelf-hosted: Docker Compose, GHCR images; Railway / AWS / Vercel guides
ModelsAnthropic and OpenAI providers
Agent frameworksPydantic AgentSpec and Cargo AI agent definitions
IntegrationsSlack, Gmail, Google Sheets, Notion, GitHub; external event triggers
IdentityGoogle, Microsoft Entra ID, generic OIDC
Open SourceYes — MIT

Strengths

  • Governance as a first-class feature — version control, code review, audit logs, identity, and RBAC are inherited from the Git workflow rather than bolted on[3]
  • True self-hosting — runs, logs, and identity stay in your environment; no vendor-hosted control plane required[1]
  • MIT license — permissive with no source-available caveats, unusual among team agent platforms
  • Non-developer authoring path — plain-language requests become reviewable PRs via the Tembo platform, so PMs can propose agent changes without writing code[3]
  • Fast iteration — 15 releases in the project's first month, calendar-versioned[1]

Cautions

  • Very early — 3 stars, 0 forks, and 34 open issues as of June 2026; no GA designation and minimal community track record[2]
  • Narrow surface coverage — Slack is the only chat surface; no Discord, Teams, or Telegram adapters
  • Two model providers — Anthropic and OpenAI only; no local model or OpenRouter support documented[1]
  • PR-gated changes cut both ways — every agent tweak requires a review cycle, which is governance for some teams and friction for others
  • Conflict-of-interest note — the author of this site is CEO of Tembo; weigh this profile accordingly

Pricing & Licensing

TierPriceIncludes
Self-hostedFreeFull product, MIT-licensed, your infrastructure and model keys

Licensing model: MIT — fully open source, no conversion clock or commercial restrictions.[2]

Hidden costs: model API usage (Anthropic/OpenAI keys are yours), hosting infrastructure, and operational time to run the Docker stack. A managed or commercial tier is not publicly disclosed.


Competitive Positioning

Direct Competitors

CompetitorDifferentiation
SpacebotSpacebot is a multi-surface (Discord/Slack/Telegram) agent runtime; Agent Studio is a Git-governed control plane
RunbearRunbear is hosted SaaS connecting LLM apps to Slack/Teams; Agent Studio is self-hosted and PR-gated
Hosted agent buildersWeb-form prompt editing with vendor-held state vs. agent definitions versioned in your own repo

When to Choose Agent Studio Over Alternatives

  • Choose Agent Studio when: agents must live in your environment, changes must be reviewable, and your team already thinks in Git workflows
  • Choose Spacebot when: you want a community-facing agent across Discord/Slack/Telegram with concurrent multi-user chat as the priority
  • Choose Runbear when: you want hosted, no-code Slack/Teams agents without running infrastructure
  • Choose hosted builders when: speed of prompt iteration matters more than governance

Ideal Customer Profile

Best fit:

  • Teams with compliance or data-residency requirements that rule out hosted agent platforms
  • Engineering-led organizations that want agent changes reviewed like code
  • Tembo platform users who want an open-source governance layer for non-coding agents
  • Teams standardizing on Slack + Google Workspace + Notion + GitHub, which map to the built-in connections

Poor fit:

  • Teams that want a hosted product with zero infrastructure to run
  • Communities needing Discord, Teams, or Telegram surfaces
  • Solo users who don't need review gates or RBAC

Viability Assessment

FactorAssessment
Financial HealthBacked by Tembo ($20M raised); Agent Studio itself is a free OSS project
Market PositionNew entrant in team agent platforms; differentiated on Git governance and self-hosting
Innovation PaceHigh — 15 releases in the first month, active pushes
Community/EcosystemMinimal so far — 3 stars, 0 forks as of June 2026
Long-term OutlookDepends on Tembo's continued investment and whether GitOps-for-agents resonates

Agent Studio is a bet that teams will want agents governed the way production code is governed. The architecture and license are sound; the open question is adoption, which at one month old is effectively zero.[2]


Bottom Line

Tembo Agent Studio is the most explicitly GitOps-shaped entry in the team agent platform category: MIT-licensed, self-hosted, with agent definitions in your repo and every change a PR.[1] It pairs naturally with Tembo's coding agent platform, which handles the authoring loop.[4]

Recommended for: governance-minded teams that want shared agents on their own infrastructure and already live in Git, Slack, and GitHub.

Not recommended for: teams wanting a hosted product, broad chat-surface coverage (no Discord/Teams), or a battle-tested community — the project is weeks old.

Outlook: promising design, unproven adoption. Watch the star count and release cadence over the next two quarters.


Research by Ry Walker Research • methodology

Disclosure: Author is CEO of Tembo.