01 · Triage
Swipe through the queue
Everything that needs attention becomes a card, with an AI summary and a read on its priority and on how safely it can be automated. Delegate it to an agent, a teammate, or yourself with a swipe.
Open core · local or self-hosted
The operations console for coding agents.
Triage Factory watches GitHub, Jira, Linear, and Slack, turns what needs attention into tasks for the team that owns it, and runs each one with an agent in an isolated sandbox, on infrastructure you control.
brew tap sky-ai-eng/tap && brew install triagefactory Triage
Triage everything that reaches your team.
Triage Factory scans everything, filters work based on fully configurable rules, and autonomously executes where needed.
How it works
From an event to a merged pull request.
You write the prompts, compose them into blueprints, and wire each one to the events that should start it. Triage Factory does the rest, and you can step in at any point.
02 · Blueprints
Wire events to blueprints
A blueprint is an ordered set of prompt steps plus the criteria that fire it. A PR opens, a label lands, a ticket moves, and the matching blueprint runs. Each step gets its own context window, and all of them share one memory.
03 · Autonomy
Set the confidence bar
Every blueprint carries a confidence threshold. An agent acts only when it can plausibly succeed; below the bar, the work waits in the queue for a person. You decide where the bar sits.
04 · Board
Follow every run on the board
Queued, in progress, done. Whatever still needs you rides the card, first in its lane, and a working run reports what it is doing as it does it, so you can open the transcript and follow along.
05 · Take over
Take over any run
Step into a running agent's workspace at any point, pick up where it left off as yourself, and hand control back when you're done.
06 · Pull requests
Every pull request in one place
Merged, draft, and awaiting-review counts, the reviews you've given and received, and every pull request still in flight, across all your repositories.
Memory
Every run starts with what your team knows.
Each team keeps a knowledge base its people write: conventions, runbooks, known problems. Every run starts with it staged in, and a note one team publishes is read by every other team's agents too. Each task also keeps a memory, so the next run on it starts from the notes the last one left.
How memory worksMulti mode
Every agent runs in its own sandbox.
Self-host Triage Factory for a team and each run gets a disposable gVisor sandbox with no usable credentials and no access to the host. Your code and your keys stay on your infrastructure.
Sandbox
No real credentials inside
The agent's environment holds placeholder tokens that point at a proxy of its own. Real keys are attached on the way out, so a dump of a live agent's environment contains nothing usable.
Inside the sandboxCapacity
Size a host by the fleet you want
A sandbox idles at about 24 MB and a few hundredths of a core, and start-up time stays flat from the 4th agent to the 128th. At full CPU saturation work slows down, and none of it fails.
Capacity, measuredInstall
Run it on your own machine.
Local mode is one binary, with its state in SQLite and your credentials in the OS keychain. No Postgres, no Docker.
brew tap sky-ai-eng/tap && brew install triagefactory