Getting started

This page gets you from a freshly deployed (or shared) TAS instance to running your first agent. If you’re standing up the instance itself, see Deploying & operating first.

1. Sign in

Open the instance URL and sign in with Google. The first time you sign in, an instance admin may need to add you to a workspace (see Audit & roles for the role model).

2. Pick a workspace

A workspace is the unit of isolation in TAS: it pins to exactly one GitHub repository and has its own members, connections, agents, and runs. Use the workspace switcher at the top of the left sidebar to move between workspaces you belong to. Instance admins can create new workspaces.

3. Connect a GitHub repository

Each workspace’s source of truth is a Git repo. On first use you’re guided to connect one — TAS stores agent definitions under agents/ in that repo and reads/writes them through the GitHub API. Once connected, the Agents list reflects what’s in the repo’s default branch.

4. Add an LLM provider key

Agents call a model, so the workspace needs at least one provider key. Go to Settings → API keys and add an Anthropic or OpenAI key. Until one is set, the sidebar shows an “LLM provider needed” prompt and runs can’t execute.

Which model?

For an agent that calls tools, start on a top-tier model (e.g. anthropic:claude-opus-4-8) to prove it works, then try downgrading to a cheaper one like anthropic:claude-sonnet-4-6 and compare cost on the Runs page. See Authoring agents for the full playbook.

5. (Optional) Authorize connections

If your agents talk to outside services (Slack, Gmail, Sheets, Attio, …), authorize them under Connections. Connections are per-user: each operator authorizes the accounts their runs act as. See Connections.

6. Create and run your first agent

• From chat: describe the agent you want; TAS opens a pull request via Tembo. Merge it and the agent appears in the Agents list.
• Run it: open the agent and use Run to execute it once. The run detail page shows the output, token usage, cost, and every tool the agent called.

From here, automate it on a schedule or an event (Automations & triggers), or refine it with Improvements.