Labs
We don't build tools. We find the structural reframe.
Every project here started the same way — not with a feature, but with a second look at the problem until its real shape showed up. Some became products. One became open source. Some are still concepts. All of them are the same habit of mind.
How we build
"Everyone was chunking code into a vector DB and losing the structure. We kept it." — from the Gristle build story
Pig Knuckle
Stop the slop.
Describe the job in plain words and get a finished result — not a response to wrangle. The problem was never the model; it was access to the expertise that makes AI deliver. We built the access layer and made the model irrelevant.
Plan the season, not just practice.
The season is the constraint that became the product. A planner that balances skill, fatigue, and safety across the whole arc — not one practice at a time.
Graphs, not chunks.
Everyone was chunking code into a vector DB and losing the structure. We kept it — a codebase mapped as a graph of calls, imports, and inheritance. 28 MCP tools. The one we gave away.
Agents that coordinate themselves.
Coordination as a graph, in production. Requests route to scored specialists that pick their own tools and feed the outcome back — our multi-agent bet, live.
Agents with careers.
The framework extracted from Ziggy — the graph is the operating system. Agents earn trust, get promoted, and spawn juniors that inherit what they learned.
Tenet 04 in action
Ziggy is the coordination bet, running live in production —
Mentat is the framework extracted from what actually held up.
Concepts & writing
The thinking is part of the work.
Here's the method underneath everything above — and the argument behind our biggest bet.