Agents as code
Every agent is a Pydantic AgentSpec file in a Git repo you own. Author by chat-to-PR or by hand — either way the change is a reviewed pull request, with draft → stable versions and a full audit trail.
Tembo Agent Studio
The self-hosted control plane for AI agents. Agents draft the work; your team signs it — every change a reviewable Git diff, every run auditable, all in your own environment.
Agents as governed production software
Section titled “Agents as governed production software”You delegate the decision, not the accountability. TAS treats agents the way you already treat code — defined in files, changed through pull requests, owned by your team, run on your own infrastructure with your own model keys — so you get outcomes you can trust and explain, not automation you have to take on faith.
Self-hosted & multi-model
Runs in your environment with your own Anthropic and OpenAI keys. Agent definitions, run history, and TAS-managed secrets stay in your infrastructure.
Humans sign the work
Match the standard to the stakes: low-stakes work runs on its own, high-stakes work routes to a shared Tasks Inbox for a person to review, correct, and approve — and every correction folds back into the agent, so the loop compounds.
Schedules, events & Slack
Cron, event triggers, webhooks, and Slack apps wired to ~1,000 Composio tools, native MCP servers, and your own secrets — with per-user auth and RBAC.
Start here
Section titled “Start here”Introduction
What TAS is and why.
Set up an instance
Run it on your own infra.
Author an agent
Write a spec, open a PR.
How is this different from Claude Managed Agents?
Section titled “How is this different from Claude Managed Agents?”Claude Managed Agents is an execution engine — an API that runs an agent in a sandbox. TAS is the control plane above execution: it decides what agents exist, how they change (Git + pull-request review), who may run them (RBAC), and how humans stay in the loop. TAS runs agents on its own bundled runner (Pydantic AI today) across Anthropic and OpenAI models, self-hosted in your environment.
How is this different from Claude Cowork?
Section titled “How is this different from Claude Cowork?”Cowork is a personal desktop coworker — single-user, tied to one machine, Claude-only, and its automations can’t be shared with teammates. TAS is the team platform: named, reusable agents that run server-side, owned by your organization, shared across coworkers, governed by review, and visible to the team in one dashboard.