XYZZY.

Multiplayer AI workspace

A team makes a hard technical decision with AI, and keeps the receipts.

Conversation branches into parallel specialists, your team picks what's worth keeping, and XYZZY synthesizes a decision brief with the evidence traced beneath it.

View on GitHub Run the demo locally
XYZZY workspace: a channel conversation alongside an open thread showing replies to one message.

The loop

One conversation, several specialists working at once, one accountable decision.

The XYZZY loop A conversation branches into three parallel specialist agents. Each output is explicitly included or excluded. What is included is combined into one synthesis, published as a decision brief with provenance. Conversation shared room Architect specialist run Researcher specialist run Security specialist run ✓ ✓ ✕ Synthesis included only Decision Brief with provenance The XYZZY loop A conversation branches into three parallel specialist agents. Each output is explicitly included or excluded. What is included is combined into one synthesis, published as a decision brief with provenance. Conversation Architect ✓ Researcher ✓ Security ✕ Synthesis · included only Decision Brief with provenance

Conversation → parallel specialist branch → select or exclude → synthesis → decision brief with provenance.

Why it's different

Governed execution, a provable audit trail, and infrastructure you control

Governed

Actions wait for human approval before they execute. What an agent may do is re-read from the room's own state at the moment it acts, so leaving a room or losing access takes effect immediately, mid-task.

Provable

Every room's event log is hash-chained: each event is hashed against the one before it, so altering or deleting a row breaks every hash after it — tamper-evident by construction, checkable with the audit CLI. Each Decision also links to the Claims and AgentOutputs behind it, so a synthesis is inspectable down to the run that produced it.

Yours

One Python process and a SQLite file, self-hosted. Point specialists at Ollama, LM Studio, or any OpenAI-compatible server instead of a hosted API. Apache-2.0 licensed, source included.

Apache-2.0 licensed. One Python process, no external services required. 857 tests plus Ruff and strict mypy, green on every push. Security posture and disclosure process in SECURITY.md.

Try it in one command

No account, no signup. It runs on your machine.

docker run -p 8000:8000 -e XYZZY_DEMO=1 ghcr.io/project-nexus-yr/xyzzy

Opens a seeded demo workspace at http://localhost:8000, signed in with one click. No account, no config. Prefer source? git clone the repo and run docker compose --profile demo up.

No Docker? It's one Python process — see the README.