# Positron AI

> Positron AI builds purpose-built Transformer inference hardware — the Atlas appliance (8x in-house
> Archer accelerators, 256 GB HBM), with Asimov silicon and the Titan system slated for 2027. Its
> Olivaw serving layer publishes two documented HTTP APIs: an OpenAI API-compatible inference
> endpoint (models, chat completions, text completions, SSE streaming) and an administrative API for
> models, service nodes, users and access tokens. Founded 2023, Reno NV; $230M Series B (Feb 2026)
> at a >$1B valuation.

Generated: 2026-08-02
Method: generated (no provider-published /llms.txt — support.positron.ai/llms.txt and
www.positron.ai/llms.txt both return 404, probed 2026-08-02)
Source: apis.yml and the artifacts in this repository

## APIs

- [Positron Olivaw OpenAI-Compatible Inference API](https://support.positron.ai/api-documentation): List models, retrieve a model, create chat completions and text completions against models loaded on Positron hardware. API key in the `authorization` header. 4 operations.
- [Positron Olivaw Admin API](https://github.com/positron-ai/admin-api-docs): Control plane for the model catalogue, service nodes (Giskard / vLLM / OpenAI backends), platform users and access tokens. Bearer API key. 19 operations.

## Specs

- [Olivaw OpenAI API — OpenAPI 3.1.0](openapi/positron-inference-openapi.yml): harvested from the published spec at support.positron.ai/api-documentation
- [Olivaw Admin API — OpenAPI 3.1.0](openapi/positron-admin-openapi.yml): harvested verbatim from positron-ai/admin-api-docs
- [Inference overlay](overlays/positron-inference-overlay.yaml)
- [Admin overlay](overlays/positron-admin-overlay.yaml)

## Docs

- [Developer portal](https://support.positron.ai/)
- [API documentation](https://support.positron.ai/api-documentation)
- [User guide / getting started](https://support.positron.ai/user-guide)
- [Release notes](https://support.positron.ai/release-notes)
- [Atlas product page](https://www.positron.ai/atlas)
- [Titan](https://www.positron.ai/titan) / [Vision](https://www.positron.ai/vision) / [Asimov](https://www.positron.ai/asimov)
- [About](https://www.positron.ai/about) / [Press](https://www.positron.ai/press) / [Careers](https://www.positron.ai/careers)
- [GitHub organization](https://github.com/positron-ai)
- [Contact sales](https://www.positron.ai/contact-sales)

## Artifacts

- [Authentication profile](authentication/positron-authentication.yml)
- [API conventions](conventions/positron-conventions.yml)
- [Error catalog](errors/positron-problem-types.yml)
- [Data model](data-model/positron-data-model.yml)
- [Lifecycle](lifecycle/positron-lifecycle.yml)
- [Changelog](changelog/positron-changelog.yml)
- [Conformance](conformance/positron-conformance.yml)
- [Packages](packages/positron-packages.yml)
- [Well-known probe results](well-known/positron-well-known.yml)
- [Domain security](security/positron-domain-security.yml)
- [Agentic access contracts](agentic-access/positron-agentic-access.yml)
- [Agent skills](skills/_index.yml)
- [MCP candidate tool list](mcp/positron-mcp.yml)

## Integration notes

- The inference API mirrors OpenAI's, so an OpenAI client library works by repointing `base_url`
  at a Positron endpoint and supplying a Positron key. Positron adds `top_k`, `seed` and
  `ignore_eos` to the selection settings.
- `functions` / `function_call` appear in the chat schema but the published document states the
  feature is not currently supported.
- Streaming is server-sent events (`text/event-stream`) via `stream: true`.
- Errors are OpenAI-style `{object: "error", type, message}` — not RFC 9457 problem+json.
- No idempotency keys, no pagination, no rate-limit headers and no published quotas.
- Positron ships no first-party SDK, no CLI, no MCP server, no webhooks/AsyncAPI, no status page,
  no public pricing and no /.well-known/ documents (all probed 2026-08-02).