New Test Run

2 runs currently active:

smoke-xsr ? · ?

smoke-ent ? · ?

You can start another run in parallel — use "Skip purge" to keep the brain intact for the active run.

Suite

Smoke — 1 case — ~5 min, pipeline health check

Targeted — ~36 cases, brain-creator curated — just what changed

Slice — 5 core cases — ~15 min

Full — All 100 cases — several hours

Parallel workers

Reps per case

Brain cleanup

Skip if another run is active.

Model Manage keys / add model →

DeepSeek

deepseek-v4-pro

deepseek-v4

deepseek-chat

Kimi (Moonshot AI)

kimi-k2.6

▸ Per-phase model overrides (optional)

Generate

No MCP — creates test data

Ingest

MCP — stores docs in brain

Query

MCP — retrieves from brain

Evaluate

No MCP — judges answer

Thinking / Reasoning

Max

DeepSeek: full chain-of-thought · Kimi: 8192 thinking tokens · OpenAI GPT-5.x: reasoning_effort=max

High

DeepSeek: high effort · Kimi: 2048 thinking tokens · OpenAI GPT-5.x: reasoning_effort=high

None

No reasoning — pure generation, fastest and cheapest for all providers

Budget control per provider — select a model above to see which applies.