Version control
Knowledge lives in git beside the repo — branchable, reviewable, yours.
In the product,
it's a wiki.
Plain Markdown in a git repo you own. Every page cites its sources.
paniolo wiki keeps it valid, and your agents search it through MCP.
Discipline you already
know from code.
Lint · consistency · provenance
paniolo wiki treats the corpus like code: page shape, links that resolve, and a cited source on every page. Rename or move a page and every link is rewritten.
Local · Rust · on demand
Local search in milliseconds, without burning a single token from your AI vendor. Agents search what they need on demand through MCP, and a hook adds relevant results automatically.
The Evolution Loop
Tasks
Real-world coding benchmarks
Trajectories
Agent execution traces
Structured Experience
Distilled root causes
Harness Mutation
Falsifiable edits with predictions
Verified Improvement
Measured, reverted if failing
The foundation of
reliable AI engineering
Built from years of production experience — before the industry named it harness engineering.
Every AI error is infrastructure debt.
When an agent makes a mistake, the right response is not to fix the output and move on. It is to update the harness so the same failure is structurally prevented. Errors are signals. Signals become infrastructure.
Harness quality is model-agnostic.
A well-engineered harness outperforms a better base model with a weak one. This is empirically validated at ICLR 2026, and it is the core premise Paniolo is built on. The infrastructure layer is the lever.
Observability before optimization.
You cannot reliably improve what you cannot see. Every component, trajectory, and decision in the harness must be auditable before any evolution loop can be trusted. Structure precedes speed.
Three pillars of
observability
Validated at ICLR 2026. The methodology that makes autonomous harness evolution reliable.
Component
Observability
Every harness component — capstone file, tool descriptions, middleware, skills, sub-agent configs, long-term memory — gets a file-level representation. The action space becomes explicit, auditable, and revertible.
Experience
Observability
Raw agent traces are distilled into a structured evidence corpus — root causes your evolving harness can act on. The intelligence layer compounds over time without requiring manual inspection.
Decision
Observability
Every harness edit is paired with a self-declared prediction, verified against next-round outcomes. Each edit becomes a falsifiable contract. Ineffective edits are rolled back automatically.