Quality modules
What each add-on module catches, the endpoint it runs on, and which tier includes it — evals through Copilot.
Evals
Clusters your production traffic, labels it with LLM-as-judge, and builds the eval sets from what users actually asked. Regressions between versions surface as a diff.
POST /api/v1/evals/runsIncluded from the Team tier, or $29 per month on its own. See Pricing.
Retrieval Monitor
Recall@K, precision, MRR and NDCG per query cluster, continuously. You hear about retrieval drift when it starts, not when support does.
POST /api/v1/retrieval/runsIncluded from the Pro tier, or $109 per month on its own. See Pricing.
Traces
Every hop of an agent run: tool calls, model calls, the decision at each step, and the latency and cost each one added.
GET /api/v1/tracesIncluded from the Starter tier, or $9 per month on its own. See Pricing.
Safety Shield
An inline layer over model output that catches PII, prompt injection, jailbreaks and hallucinated claims — and blocks, redacts or flags them.
POST /api/v1/safety/classifyIncluded from the Pro tier, or $109 per month on its own. See Pricing.
Cost Attribution
Spend broken down by feature, user, prompt template and provider, so the bill points at a line of your product rather than at a month.
GET /api/v1/cost/summaryIncluded from the Team tier, or $29 per month on its own. See Pricing.
Migrate
Moves an embedding model, a retriever or a vector store with no downtime, and proves quality held before it cuts over.
POST /api/v1/migrate/runsIncluded from the Pro tier, or $109 per month on its own. See Pricing.
Copilot
A meta-agent on the Cloudflare Agents SDK that watches every module you have switched on. It sends a daily digest of the anomalies it found, and it acts on them — opening a pull request with the fix, proposing a migration, or capping a runaway spend before it becomes a bill.
POST /api/v1/copilot/runsIncluded from the Scale tier, or $399 per month on its own. See Pricing.
Playground
Compare two prompt configurations live, side by side, with comparative metrics.
POST /api/v1/playground/runsIncluded from the Starter tier, or $9 per month on its own. See Pricing.
Explorer
Run AI Search queries live over your production content.
POST /api/v1/explorer/searchIncluded from the Team tier, or $29 per month on its own. See Pricing.
Voice
Call volume, latency and error rate for every flow using the Voice block, with Call Replay to see the whole request one call ran inside.
GET /api/v1/voice/callsIncluded from the Scale tier, or $399 per month on its own. See Pricing.
Vision
Call volume, latency and error rate for every flow using the Vision block.
GET /api/v1/vision/callsIncluded from the Scale tier, or $399 per month on its own. See Pricing.
Custom Script
A Pro+ escape hatch for logic the plain transform block cannot express — a decision or a computed value, not just a reshaped field. It never executes code: a script is a JSON expression tree over a closed, whitelisted operator vocabulary (arithmetic, comparison, boolean logic, string ops, a ternary), so there is nothing to sandbox against because there is nothing outside the vocabulary a tree can reach. Used inline as the custom.script block, not called as its own endpoint.