Fullhouse  turns issues into tested draft PRs with Claude Code and Codex  ·  Install
Security

How Fullhouse handles your code and credentials.

Fullhouse is a self-hosted engine in beta. This page says what it does, what it does not do, and how to report a problem.

Self-hosted engine

The fullhouse CLI runs jobs on your machine, your server or your container. No Fullhouse cloud service sits between your agents and your code.

What leaves your machine

Your agent CLIs send prompts and code to their own providers, as they do without Fullhouse. The GitHub CLI pushes branches and opens draft PRs. Fullhouse itself sends no telemetry.

Credentials

Fullhouse never asks for provider API keys. It uses the logins your agent CLIs and the GitHub CLI already have and does not copy them into its own storage. The optional remote HTTP server uses a bearer token you create yourself.

Admission and review

Only issues labeled by someone with write access to the repository are admitted. By default each job works in a private clone on its own branch and ends in a draft PR. Nothing merges automatically.

Redaction

Transcripts are redacted before a handoff, and the handoff is refused if anything still looks like a secret. Redaction is defense in depth, not a guarantee: keep secrets out of issues and repositories.

MCP access

The stdio server grants read and write by default; starting jobs needs the execute scope. The remote HTTP server (fullhouse mcp serve) requires a bearer token and binds to 127.0.0.1 by default. Keep it on localhost or behind your own TLS proxy, and do not expose it to the internet.

Integrity, stated plainly

Evidence files carry a SHA-256 digest, and MCP receipts form a hash chain, so later edits are detectable. Neither is a cryptographic signature, and builds are not code-signed yet. Check downloads against the published SHA-256 checksums.

Report a vulnerability

Email hello@mosaiclab.net with a description, steps to reproduce and the impact. Please do not report security issues in public. You should hear back within a few business days. We ask for a 90-day disclosure window from the first report so we can investigate, fix and release.

Last updated: October 18, 2026