Workflows
Build n8n-style pipelines where the nodes are models: one drafts, a council of AIs reviews the work, issues get fixed automatically, and nothing ships without consensus. The builder below is a live simulation.
Trigger
New pull request / task submitted
Draft
Writes the first version
claude-opus-4-1
Anthropic reviewer
gpt-5.2
OpenAI reviewer
gemini-3-pro
Google reviewer
Consensus
Merges verdicts · majority rules
Auto-fix
gpt-5-mini applies the council's notes
fix ×1Validate & ship
Final check, then response + ledger
# Press “Run pipeline” — or click nodes: reviewers can be benched, the trigger runs, auto-fix toggles.
A council, not a single judge
Three different frontier models review the same draft in parallel. Disagreement is the feature: what one model misses, another flags, and majority rules.
Issues fixed in the loop
When the council requests changes, a fast fixer model applies every note and sends the revision back for a re-vote, automatically, until consensus or your retry cap.
Every verdict on the ledger
Each draft, review, fix and re-vote is a traced call with its exact cost, so you can see precisely what quality assurance costs, per task.
Why a council
Different models, different blind spots
Self-review is flattery: a model grading its own work misses its own failure modes. A cross-vendor council catches what monocultures can't, and because it all runs through the gateway, budgets, locks and user-assigned keys apply to every seat at the table, including your local models via the Agent.
Ship this pipeline for real
The visual builder lands in the console; the same council is one JSON policy on any API key today.