Skip to content

Glossary

A quick reference for the terms you’ll meet across the BatchRouter API and docs. Definitions are one line each; follow the linked guide or the API reference for full field details.

API key — Your bearer credential, prefixed br_live_, sent as Authorization: Bearer br_live_…. Keys are shown once at creation; create them in the dashboard or via POST /v1/auth/account/api-keys (or agent-first with POST /v1/auth/agent-register). See Authentication.

Artifact — The assembled output (or error) file for a completed batch, stored as JSONL. Fetch a signed, time-limited download link with GET /v1/batches/{batchId}/artifact-url, or read items inline via GET /v1/batches/{batchId}/results. See Retrieve results.

Batch — A job you submit once with POST /v1/batches; it gets a bat_… id and moves through the status lifecycle (pending → queued → routing → dispatched → processing → completing → completed, or terminal failed | cancelled | expired). One batch can fan out into several internal lanes. See Submit a batch.

Billing receipt — The per-batch settlement record (what you were charged, against which credits). Fetch it with GET /v1/batches/{batchId}/billing-receipt, or embed it in the batch response with ?include_billing_receipt=true. See Credits and billing.

Completion window — The deadline BatchRouter routes against, derived from your SLA tier (standard ≈ 24h). The OpenAI-compatible surface exposes it as the completion_window field.

Credit — Prepaid balance (a micro-USD wallet) that batch creation draws down. Buy credits in the dashboard — there is no public billing-checkout endpoint. See Credits and billing.

customer_item_id — Your own identifier on each input item; BatchRouter echoes it back on the corresponding result so you can join outputs to inputs. See Submit a batch.

Delivery target — A registered customer-owned, S3-compatible bucket (AWS S3, Cloudflare R2, GCS S3-interop, MinIO) that BatchRouter pushes result and error artifacts to on finalize. Manage via GET|POST /v1/delivery-targets and test connectivity with POST /v1/delivery-targets/{id}/test. See Delivery.

Idempotency key — The Idempotency-Key header you send with POST /v1/batches so a retried request creates the batch once, not twice. See Submit a batch.

Item — One request inside a batch (e.g. a responses or embeddings call). The canonical item carries a customer_item_id, operation, model, and input. List them with GET /v1/batches/{batchId}/items.

Lane — An internal execution path (one provider offering for one model) that BatchRouter selects to run part or all of your batch. A batch with mixed-model items splits into model-specific lanes; non-selected lanes carry a rejection receipt. See How routing works.

Model catalog — The public list of models you can request, with metadata and pricing inputs. Browse with GET /v1/catalog/models and GET /v1/catalog/models/{slug}. Always source real model slugs from the catalog rather than hardcoding them.

Operation — The task type of an item — for example responses or embeddings. It tells BatchRouter how to shape and route the request. See Submit a batch.

Privacy tier — How strictly a batch is routed on data handling: standard (default), confidential (providers with data-retention opt-out), or restricted (private BatchProviderApi nodes only). See How routing works.

Provider — A third party that runs the actual inference on a lane. Browse the marketplace with GET /v1/providers and GET /v1/providers/{slug}. BatchRouter owns routing, pricing, and billing; providers own their infrastructure.

Quote (quote_id / qlock_) — A free cost estimate from POST /v1/quotes/model (or POST /v1/quotes/workflow). It returns a quote_id shaped qlock_… — a price-locked snapshot you pass to POST /v1/batches to reserve credits at the quoted price before dispatch. See Quotes.

Rejection receipt — The auditable record on each non-selected or ineligible lane explaining why it wasn’t routed: a normalized rejection_code (e.g. capacity_full, stale_heartbeat, privacy_tier_mismatch, region_unavailable), the failed eligibility checks, and provider/offer details. This is what makes BatchRouter routing explainable. See How routing works.

Retention — How long result artifacts are kept. Defaults to free grace_only; extended_fixed (metered) keeps them longer, and never_charge guarantees zero storage spend. Manage with GET|PUT /v1/retention-settings. See Delivery.

Routing mode — How BatchRouter ranks eligible lanes: cheapest (default), sla_aware, public_only, edge_only, hybrid, or privacy_constrained. See How routing works.

SLA tier — The speed/price class for a batch: standard (≈24h SLA, default), flex (up to 48h, lower price), or priority (expedited). It sets the completion window routing targets. See How routing works.

Webhook — An HTTPS callback BatchRouter sends on batch events, signed with X-BatchRouter-Signature (HMAC-SHA256). Set per-batch ({ url, secret }) or an org default via PUT /v1/auth/account/delivery-webhook; inspect delivery with GET /v1/batches/{batchId}/webhooks. See Webhooks.

Workflow product — A curated outcome contract — a predefined manifest (input schema, output schema, SLA, route policy) for an agent-native task, not a general orchestration engine. Quote one with POST /v1/quotes/workflow; browse with GET /v1/catalog/workflow-products. See Workflow products.