Execution runtime
LLM work, as an API primitive.
Orkiva runs reusable Skills and goal-driven Loops against your own provider keys. Your app calls POST /v1/runs. We handle iteration, evaluators, stop conditions, tokens, and cost.
RUN #18472 · Prompt Optimizer
Running · iteration 4 / 10
91 / 95
- Current cost
- $0.087
- Est. maximum
- $0.214
- ✓ Iteration 172 · $0.018
- ✓ Iteration 281 · $0.021
- ✓ Iteration 387 · $0.024
- ● Iteration 491 · $0.024
- ○ Iteration 5–10
Skill
A reusable capability: prompt, schema, model, evaluator.
Loop
Iterate until score ≥ goal, or cost / time / iterations stop it.
Run
One execution. Tokens, USD, credits, events — live.
01
Bring your own key
OpenAI, Anthropic, OpenRouter. Credentials stay encrypted in Orkiva. Applications never see the provider secret — they hold an ork_ key.
02
Goal, not chat
A Loop is not a conversation. It is a process: generate, evaluate, decide. Threshold, interval, max cost, timeout. When the score clears the goal, it stops.
03
Blind verification
Datasets split into ground truth, evaluation, and a locked blind set. Optimization cannot see the blind examples. The report shows the generalization gap.