Docs
API Reference

API Reference Overview

Base URL, authentication, surfaces, and the full endpoint map for the Ablatic API

This page is the map of the Ablatic API: the base URL, how to authenticate, the two compatible surfaces, and every public endpoint.

Base URL

https://api.ablatic.ai/v1

The host api.ablatic.ai proxies only /v1/* and /health. Any other path returns 404.

Authentication

Every request except the unauthenticated probes below needs an API key. Keys look like sk-ablatic-... and are created in the console. A key is shown once at creation and stored only as a SHA-256 hash, so it cannot be recovered.

Two headers carry the key, one per surface. Both run the same validation.

Authorization: Bearer sk-ablatic-...
x-api-key: sk-ablatic-...

Use Authorization: Bearer on the OpenAI-style endpoints. Use x-api-key on the Anthropic-style endpoints.

The surfaces

The API exposes the same model over three request and response shapes. Pick the one your client speaks; the model and pipeline behind them are identical.

  • OpenAI Chat. Drop-in for the openai SDK. Base URL https://api.ablatic.ai/v1, endpoint /v1/chat/completions, model talos, Authorization: Bearer.
  • OpenAI Responses. For clients that default to the Responses API (Vercel AI SDK, Codex-style agents). Endpoint /v1/responses, stateless only. See Responses.
  • Anthropic Messages. Drop-in for the anthropic SDK and Claude Code. Base URL https://api.ablatic.ai, endpoint /v1/messages, key via x-api-key.

The default model is talos. You select the reasoning mode with reasoning_effort. See Models.

Endpoints

OpenAI-style

MethodPathPurpose
POST/v1/chat/completionsChat completion, streaming SSE
POST/v1/responsesResponses API completion (stateless)
GET/v1/modelsList the model (no auth)
GET/v1/auditList Glassbox audit records
GET/v1/audit/statsAudit summary stats
GET/v1/audit/{request_id}Fetch one audit record
DELETE/v1/audit/{request_id}GDPR erasure of one record
POST/v1/filesUpload a file for context injection
GET/v1/filesList uploaded files
GET/v1/files/{file_id}Fetch one file
DELETE/v1/files/{file_id}Delete one file
POST/v1/agent/runServer-side autonomous agent loop
POST/v1/agent/run/{run_id}/cancelCancel a running agent
POST/v1/simulation/analyzeAnalyze HDF5, NumPy, or video data

Anthropic-style

MethodPathPurpose
POST/v1/messagesAnthropic Messages completion
POST/v1/messages/count_tokensEstimate input tokens (heuristic)
GET/v1/messages/_healthHealth probe (no auth)
POST/v1/messages/batchesCreate a message batch
GET/v1/messages/batchesList batches
GET/v1/messages/batches/{id}Fetch one batch
POST/v1/messages/batches/{id}/cancelCancel a batch
DELETE/v1/messages/batches/{id}Delete a batch
GET/v1/messages/batches/{id}/resultsStream batch results (NDJSON)

Message Batches run on the Anthropic-compatible surface. The OpenAI-style /v1/batches endpoint is not offered. See Batch.

No authentication

Three endpoints need no key: GET /v1/models, GET /v1/messages/_health, and GET /health.

Common headers

These headers ride on most requests and responses across both surfaces.

DirectionHeaderMeaning
RequestX-TimezoneTimezone for usage and billing windows. Default UTC
RequestIdempotency-KeySafe-retry key on batch creation
ResponseX-Request-IdUnique id for the request, quote it in support tickets
ResponseX-RateLimit-LimitConcurrent-request limit (see note below)
ResponseX-RateLimit-RemainingRemaining capacity
ResponseX-RateLimit-ResetWhen the window resets
ResponseRetry-AfterSeconds to wait, sent on 429 and 503

X-RateLimit-Limit currently reflects the concurrent-request limit, not requests per minute. Do not read it as RPM. See Rate limits.

Next