One control plane for every agent loop.
FastAssert coordinates, verifies, and audits the work of AI agents — whichever vendor’s model, whoever’s compute. We run zero inference and zero tenant code; we run the control room.
Control-plane capabilities
Coordinate, verify, and audit — in one place.
Verification
Gates
Pluggable verification: command exit, JSON schema, webhook, metric threshold, and human approval. They compose — all must pass before a task is accepted.
Visibility
The board
Workspaces hold loops and epics→tasks with dependencies. Every loop carries a versioned plan the agent re-reads each run — and you watch it all move in real time.
Coordination
Collision-free leasing
Tasks are claimed through atomic, time-boxed leases, so two agents can never grab the same work — and if a run dies, its lease expires and the task returns to the board automatically.
Durability
Hosted triggers
Cron schedules fire from the control plane, laptops closed. When no runner is live, it raises an alert — never silence.
Any vendor
Bring your own everything
Claude, Codex, Cursor, or your own script — any agent CLI flows onto your board. Their agents, their tokens, their compute; our coordination. We run zero inference.
The full control-plane surface
Nine capabilities, one control plane.
Live fleet board & heartbeats
Every loop, epic, and task on one board, with runner heartbeats marked live, stale, or dead — the same view from the dashboard or your agent's MCP tools.
Loop template gallery
One-click starter loops — pr-babysitter, issue-triage, dep-watch, staging-validation — each shipping with instructions, default gates, a schedule, and its state shape.
Schema-validated journal
Every run writes an append-only journal record: outcome, model, tokens in and out with cache split, actions, and artifacts. Malformed rows are rejected at write time.
Model policy & token budgets
Set the model per loop or task; we pass it to your agent CLI and never route a token. Per-loop budgets stop scheduling before you overspend — and because the runner handles coordination itself, your agents spend no tokens talking to us.
Atomic task leasing
Claims are atomic and time-boxed, renewed by heartbeat while the agent works, and released with an outcome. Two agents never grab the same work, and a crashed run frees its task automatically.
Hosted cron triggers
Schedules fire from the control plane with every laptop closed. Runners pick up work on poll, and a 'no live runner' condition raises an alert — it never fails silently.
Human approvals & multi-seat
Judgment calls route to a dashboard approval queue; approve or reject with a comment, recorded in the audit chain. Multi-seat approval comes with the Team tier.
Hash-chained audit log
A per-org, append-only ledger where each record is sealed against the one before it. Any run can be reconstructed end to end, and tampering detectably breaks the chain.
Composable gates
Command exit codes, JSON schema checks, webhooks, metric thresholds, and human approval. Gates compose, and all of them must pass before a task is accepted.
Scope
FastAssert is the control room, not the factory. It is not an execution platform — your code never runs on our infrastructure. Not an inference provider — zero LLM calls in the product path. And it is not a CI/CD replacement, an agent framework, or a prompt tool. Your agents, your tokens, your compute — our coordination.
Install the runner in four lines.
curl, log in, install the runner, and watch it turn green in the dashboard. Coordinate, verify, and audit your fleet from day one.