Skip to main content

How tapagents works

tapagents is a calm executive view for the AI agent team you already run on top of Claude Code — fitted at start, current over time, and visible at every consequential decision.

The cold start

Sign in with GitHub. Pick a project name and stage. The Tailor-your-team wizard reads your project type and risk surface, then scaffolds the agent roster that fits — no useless agents, no over-broad swarm. Your first scaffold ships as a pull request in the new repository within your first session.

What your team does

Decision Packets

Every consequential decision lands as a structured packet — citations, dissent, severity, your verbatim approval. Replayable, exportable, accountable.

State machine

Each project carries a phase, a blocker count, a pending-decision count. No hand-rolled status fields; the dashboard reads the machine.

Dissent log

When agents disagree, the dissent surfaces — not hidden in scrollback. You see what was considered before the chosen path landed.

Live telemetry

Tap Agents Live tails every agent run. Watch decisions land, catch issues before they ship, replay the whole thing.

The flow end-to-end

  1. 1

    Sign in

    Connect GitHub. The dashboard issues a single-use session cookie scoped to tapagents.ai.

  2. 2

    Tailor your team

    Pick project type, stage, and risk surface. The wizard scaffolds the right agent roster.

  3. 3

    Run agents

    Work in Claude Code as you normally would. The team reads the scaffold and routes your prompts to the right specialist.

  4. 4

    See live activity

    Tail every agent run from the dashboard. Decisions, dissent, citations — all surfaced as they land.

  5. 5

    Approve packets

    Consequential decisions wait for your verbatim approval. Replayable, exportable.

  6. 6

    Ship

    Each project's state machine carries it forward. You scan the dashboard; the dashboard reads the machine.

Where it runs

Claude Code runs on your machine. tapagents is the dashboard you check to see what your team is doing — agents execute locally; the executive view surfaces telemetry, decisions, and packets through the GitHub App OAuth flow you already trust. No server-side agent execution, no shadow-cloud running on your behalf.

Ready to try this?

Sign in with GitHub →

Free during beta.

How it works | TapAgents