Open source

Projects I build in the open: real software with public code, honest status notes, and no telemetry.

The guides on this site describe how agent systems should behave. These projects are where I hold myself to it: real software, public code, and a status section that says plainly what has not been tested yet.

More are coming. Each one ships with its source, its licence, and a link back to the writing that shaped it.

On the bench

3 projects · updated Nov 2026

RunningSince Aug 2026
Project 01 · Agent platform

yaaf

Run a team of AI agents like a company.

Describe a goal. yaaf hires a CEO and specialist teams, gives them a budget and a deadline, and keeps them working for weeks while you sit as chairman. The rules live in the server, not the prompt: an agent can’t mark its own work done, overspend, or quietly move the date.

  • “Done” means the platform ran the tests itself
  • Budgets that pause the project at the limit
  • Any model per seat; runs as an MCP server inside Claude
  • Rehearse a whole project for $0 on scripted agents
PythonApache-2.0MCPSelf-hosted

Status

✓Running real projects end to end
✓Claude Code, Codex, Gemini CLI, OpenCode
✓Free rehearsal mode
v0.1Pre-release
Project 02 · Desktop app

Rowboat

A fast, lightweight SQL client with an AI copilot that knows your schema.

For macOS, Windows and Linux. PostgreSQL, ClickHouse and SQLite, with a real editor and data grid and nothing that slows startup. The copilot writes, fixes and explains SQL, but AI-generated SQL never runs until you press Run, and your rows are never sent to a model.

  • Cold start under 1 s, idle memory under 150 MB
  • Schema-aware autocomplete: aliases, CTEs, foreign-key joins
  • Secrets in the OS keychain; TLS and SSH tunnels
  • Any LLM provider, or a local model with no key
RustMITPostgresClickHouseSQLite

Status

✓Feature-complete for v0.1
✓Built and tested on Linux
○macOS and Windows builds pending
○No installers yet; build from source
Livedemoriff.com
Project 03 · Web app

DemoRiff

Studio-grade product demos from a rough screen recording.

Turn one raw screen recording into a directed demo video and a step-by-step written guide with screenshots pulled from the footage. You cut the video by editing the transcript: delete a word and those milliseconds are gone.

  • Record or import, transcribe, edit as text
  • Render an MP4 with FFmpeg, publish to a share page or embed
  • Written guide with screenshots taken from the footage
  • Declines to show any number it can’t measure
TypeScriptNext.jsFFmpegPostgres

Status

✓Record → edit → render → publish works end to end
✓Self-host with one Postgres database
○Interactive tours and social cuts designed, not built

Next on the bench

More projects are moving to open source. Follow on GitHub to see them land, or subscribe to the RSS feed to get the write-up that goes with each one.