ModelRig — the operating layer for AI workstreams

Point your code at ModelRig and every model call becomes a route you can see, prove, and save on — while the registry keeps the routing honest: providers declare capabilities, ModelRig measures them, and a probed fact beats a declared claim.

• See — every run lands in an inspectable artifact store (runs → steps → artifacts, lineaged and hashed), with a grade and cost on each. When someone asks why the pipeline did that, you pull up the run.
• Prove — replay your own recent traffic against cheaper candidates in a bake-off and get the effective cost of conformance for each, with confidence intervals.
• Save — a winning candidate is a change to your YAML you review; nothing swaps a model on its own, and your routes are always in your git.

Start here:

• Quickstart — Step 0, then point your coding agent at it or set it up by hand; your first rig.run() in minutes.
• Observe a pipeline — see your existing runs first, without migrating any routes.
• Route bundles — the routes-as-code reference.
• Probes — the measurement harness: schema conformance (parses ≠ conforms ≠ values-correct), grounding, caching, image input.
• Leaderboard — models ranked by conformance per dollar, declared-vs-probed discrepancies flagged.

Bring your own provider keys (they stay in your environment; the first 1M requests each month are free, then a flat 2% of list price) or route on managed keys (ModelRig fronts the cost — provider cost + 2%). Same flat 2% either way.

Reproduce any number

npx modelrig-probes run --model deepseek/deepseek-chat --class schema

Every published result carries raw samples, fixture hashes, and a 95% confidence interval. A reproduction passes when your rerun lands inside the interval. Disagreements are contributions — file an issue.