A durable software factory for agents

Agent work that survives the real world.

Runyard runs long-lived AI work on your machines — with recovery, human decision gates, and a complete record of every attempt.

Install Runyard Open the hub Read docs

Dispatch board Every attempt on the record

run_7232192cimplement · attempt 1 running

run_8f12b398release · decision gate waiting on you

run_c4d90a11migrate · attempt 2 recovering forward

run_2ab7e6f0audit · attempt 3, immutable complete

21:07:12 run_7232192c · step apply-migration started on runner atelier-01 21:07:40 run_8f12b398 · decision gate raised — answer on web, API, or Telegram 21:08:02 run_c4d90a11 · attempt 1 preserved, attempt 2 dispatched from checkpoint 21:08:19 run_2ab7e6f0 · verified — logs, diff, and usage attached to the record

The hub is the ledger

One append-only record of everything your agents do.

Launch work, answer decisions, and inspect the evidence in one place — nothing vanishes when a process dies or a laptop sleeps.

Stable runs, immutable attempts

Retries never rewrite history: one run ID, every attempt preserved with its own logs, diff, and outcome.

run_c4d90a11 · attempt #1 failed · attempt #2 succeeded — both on the record

Decision gates that wait, recovery that resumes

Approvals suspend work without failing it — answer from web, API, MCP, or Telegram and the same run continues. Crashes re-park; nothing restarts silently.

gate needs you → approve · request changes · reject → the same run resumes

Evidence by default

Logs, artifacts, diffs, token usage, budgets, and every human decision stay attached to the run that produced them.

logs · artifacts · +182 −46 · 12k tok · decision: approved by fran

A copilot and a fleet, under your control

The Copilot shelf explains state and proposes confirmation-gated actions; runner fleets report capacity, queues, and sandboxes across your machines.

atelier-01 online · 2 slots · sandboxed · scoped tokens · audit log

Work/Factory

From signal to reviewed implementation on one rail.

The Work board connects the request, the approved direction, the execution, and the release evidence — every hand-off is a recorded decision.

Signal

Capture the request or alert as durable work.

Direction

Approve scope and risk at a gate.

Implementation

Run it, every attempt linked to the ticket.

Evidence

Review proof before promotion ships.

Interfaces

One control plane, many entry points.

Web, CLI, API, MCP, ACP, schedules, hooks, and CI all write to the same durable record. Self-hosted, one deployment per org.

web
cli
api
mcp
acp
schedules
hooks
ci
telegram

curl -fsSL https://raw.githubusercontent.com/yolo-maxi/runyard/main/install.sh | bash

runyard login --url https://hub.example.com
runyard token-create runner-local --scopes runner
runyard runner setup --workspace ~/runyard-runner --location local
runyard runner start --workspace ~/runyard-runner --location local

runyard run hello --where local \
  --input '{"topic":"durable agent operations"}'