# YATFA

> YATFA is an engineering organization, delivered as software. It works around the AI coding agents a team already uses and keeps going between their sessions: it finds work in the codebase and proposes it with reasons, cuts a roadmap direction into tickets one slice at a time, has every change read by an agent that didn't write it and then used by QA with proof attached, keeps project knowledge in versioned articles, and stops with the reason when work needs a person. How far it goes on its own is a per-project setting.

Last updated: 2026-10-06. Website: https://yatfa.com

Key facts:

- Category: an engineering organization around AI coding agents. It is not an editor assistant, not a single agent and not a prompt loop. A team can keep its editor assistant and connect it to YATFA over MCP.
- Orchestration runs without an LLM: the same situation leads to the same next step, by rules you can read.
- Work it finds on its own: between requests, YATFA reads the code, its git history, its own notes and CI, and files what it finds as proposals with reasons. Your rules decide which go ahead.
- Every change is written on its own branch, reviewed by a different agent that leaves a written verdict, then checked by QA against what the ticket promised: screenshots for a screen, real requests and output for everything else. If the review fails, the change goes back to the same ticket with the verdict attached.
- Autonomy is set per project with four switches, all off by default: approve larger proposals, confirm its own tickets, approve finished work, cancel tickets. The limits are enforced by the server, so an instruction to a model cannot talk its way past them. Small, safe work (tests, docs, small fixes, small refactors) is approved by an agent at every setting; operational changes always wait for a person.
- When work is stuck, it does not loop: it stops, records the reason and every attempt, and hands the work to a person. The work continues from the person's answer.
- Knowledge: findings from tickets, reviews and QA are folded into versioned articles, corrected when the code changes, and given to agents as summaries first.
- Access to code: a GitHub App installed on the repositories you choose. It writes branches, commits, pull requests and comments, and can be revoked in one click.
- Where agents run: on Cloud slots that YATFA operates, or on your own laptop, server or cloud VM (started with `npx yatfa worker`). Each agent works in its own container with a fresh clone. Hosting on your own machines is never billed.
- Pricing: register and you are in a working free organization, a slice of the product; paying unlocks the full product. Model compute is paid in credits, or with your own model key. Cloud capacity is billed by measured use above a free allowance.

## The work

- [Self-found work](https://yatfa.com/platform/self-found-work): Proposals YATFA files on its own from the code, git history, notes and CI, each with its reasons; your rules decide which proceed.
- [Roadmaps](https://yatfa.com/platform/roadmaps): Describe a direction instead of tasks; YATFA cuts it into tickets one slice at a time, each from the codebase as it stands after the last slice landed.
- [Visual drafts](https://yatfa.com/platform/visual-drafts): Several interface directions rendered from your running product before anything is built; the one you pick becomes the work.
- [Automated QA](https://yatfa.com/platform/automated-qa): After review, QA works through what the ticket promised and attaches proof to the ticket.
- [Code review](https://yatfa.com/platform/code-review): Every change is read by an agent that didn't write it, against the ticket, with a verdict in words.

## The controls

- [Autonomy](https://yatfa.com/platform/autonomy): Four per-project switches decide what YATFA may start, approve and finish; limits are enforced by the server and every step is recorded.
- [Escalations](https://yatfa.com/platform/escalations): Work that needs a person stops with the reason and the history of attempts.
- [Knowledge](https://yatfa.com/platform/knowledge): Versioned articles built from the work itself, kept current as the code changes.
- [Investigations](https://yatfa.com/platform/investigations): Ask a question about your system and get a checked report with evidence, plus a written account of every day.

## Connect

- [Web app](https://yatfa.com/platform/web-app): What agents are doing, what waits for you, what was learned and what shipped, in the browser.
- [CLI](https://yatfa.com/platform/cli): One npm command starts an agent on any machine you own; the same CLI lists, pauses and restarts agents.
- [MCP](https://yatfa.com/platform/mcp): Connect your own AI client to the workspace over MCP, within your account's permissions; its work goes through the same review, QA and rules.
- [GitHub App](https://yatfa.com/platform/github-app): What YATFA can do in your repositories, and what it cannot reach.
- [Self-hosted runners](https://yatfa.com/platform/self-host): Agents on your machines, one container per agent; the service keeps the record of the work.

## Solutions

- [For founders](https://yatfa.com/for/founders): Review, QA, current knowledge and a plan that follows the code for a small team, working overnight.
- [For engineering leaders](https://yatfa.com/for/engineering-leaders): Set direction for the organization, see which directions are moving from what shipped, read a daily account.
- [Cloud](https://yatfa.com/solutions/cloud): Agents on Cloud slots YATFA operates, billed by measured use.
- [Your own machines](https://yatfa.com/solutions/own-machines): Agents on hardware you already pay for, same rules, hosting never billed.

## Why YATFA

- [Why YATFA](https://yatfa.com/why-yatfa): What persists between agent sessions, what YATFA is not, and how it works from the outside.
- [All comparisons](https://yatfa.com/compare): Where YATFA sits among agent tools, with primary sources and where others are stronger.
- [YATFA and Devin](https://yatfa.com/compare/devin): A session-based agent and an organization; every claim about Devin links to Cognition's own sources.
- [YATFA and Factory](https://yatfa.com/compare/factory): Two ways past the single agent; every claim about Factory links to Factory's own sources.
- [Trust and security](https://yatfa.com/trust): What agents can reach, where code goes, what is recorded, and what is not in place yet.

## Pricing

- [Pricing](https://yatfa.com/pricing): A free slice of the product, the full product when you pay, credits for model compute, Cloud capacity by measured use, own machines never billed for hosting.

## Optional

- [Platform overview](https://yatfa.com/platform): Every part of the platform on one page.
- [About](https://yatfa.com/about): YATFA is built by its own organization.
- [Product updates](https://yatfa.com/blog/category/Product): Monthly posts on what shipped in the product.
- [Blog](https://yatfa.com/blog): Notes from the work.
- [Glossary](https://yatfa.com/glossary): Plain definitions of agentic engineering terms.
- [Contact sales](https://yatfa.com/contact-sales): Tell us how your team works and what you would hand over first.
- [Open source](https://yatfa.com/open-source): Warden, an open-source desktop app for watching and steering AI agents across machines.
- [Privacy](https://yatfa.com/privacy)
- [Terms](https://yatfa.com/terms)
- [Data Processing Addendum](https://yatfa.com/dpa)