# Heimdallm

> An inference marketplace: independent providers serve models behind one OpenAI-compatible API, each at their own published prices, billed on measured usage.

## Start here

- [Agents](https://staging.heimdallm.com/AGENTS.md): what the API is, how a key works, how these pages are served as markdown, and how to get a key without an account.
- [Agent skill](https://staging.heimdallm.com/SKILL.md): versioned CLI workflows for customers and providers.
- [Documentation](https://staging.heimdallm.com/docs.md): One OpenAI-compatible API for many inference providers.

## Documentation

- [Agent setup](https://staging.heimdallm.com/docs/agents.md): Install the versioned Heimdallm CLI skill and connect an agent using existing login methods.
- [Authentication](https://staging.heimdallm.com/docs/authentication.md): API keys, how they are scoped, and what happens when one is refused.
- [CLI reference](https://staging.heimdallm.com/docs/cli.md): Every `heimdall` command, its flags and its arguments, generated from the CLI source.
- [How Heimdallm works](https://staging.heimdallm.com/docs/concepts.md): The marketplace in one page: providers, endpoints, models and families; declared versus measured; and the life of a request.
- [Errors & rate limits](https://staging.heimdallm.com/docs/errors.md): The error envelope, every status the API returns, and how rate limiting works.
- [Glossary](https://staging.heimdallm.com/docs/glossary.md): Every term the marketplace uses, one line each.
- [MCP server](https://staging.heimdallm.com/docs/mcp.md): Connect an agent to Heimdallm over the Model Context Protocol: model discovery, inference, media, files, batches, keys and provider operations as tools.
- [Switching to Heimdallm](https://staging.heimdallm.com/docs/migration.md): Move an existing OpenAI, OpenRouter or OpenAI-compatible integration to Heimdallm: what transfers untouched, and what differs.
- [Models & routing](https://staging.heimdallm.com/docs/models-and-routing.md): Model ids, families, routing suffixes, fallback models and provider preferences.
- [Pricing & billing](https://staging.heimdallm.com/docs/pricing.md): How requests are priced, what usage.cost means, and how credits and spend limits work.
- [Quickstart](https://staging.heimdallm.com/docs/quickstart.md): Sign in, create a key, make your first request.
- [Heimdallm SDKs](https://staging.heimdallm.com/docs/sdk.md): The TypeScript and Python clients of the whole API: inference in every dialect, media jobs, files, batches, the catalogue, the account, keys and provider routes.

## API reference

- [Key & activity](https://staging.heimdallm.com/docs/api/account.md): GET /key, /credits, /generation and /activity — inspect the key you call with and what it spent.
- [Batch API](https://staging.heimdallm.com/docs/api/batches.md): Submit requests together, let them run in the background, and read the outcome of each from a results file.
- [Chat completions](https://staging.heimdallm.com/docs/api/chat.md): POST /chat/completions — request parameters, response fields and the streaming frame format.
- [Completions](https://staging.heimdallm.com/docs/api/completions.md): POST /completions — the OpenAI text-completions shape, kept for SDK compatibility.
- [Count message tokens](https://staging.heimdallm.com/docs/api/count-tokens.md): POST /messages/count_tokens — estimate Messages input size without running inference.
- [Decisions](https://staging.heimdallm.com/docs/api/decisions.md): POST /systemone — answer typed questions about supplied content with probabilities.
- [Embeddings](https://staging.heimdallm.com/docs/api/embeddings.md): POST /embeddings — turn text into vectors with any embeddings model on the marketplace.
- [Files API](https://staging.heimdallm.com/docs/api/files.md): POST, GET and DELETE /files — store a file once on your account and address it by id afterwards.
- [Image generation](https://staging.heimdallm.com/docs/api/images.md): POST /images — turn a prompt into images with any image model on the marketplace, buffered or streamed, with the Image API listings and their pricing lines.
- [Provisioning API](https://staging.heimdallm.com/docs/api/keys.md): GET, POST, PATCH and DELETE /keys — create, list, edit and revoke keys with a management key.
- [Mesh generation](https://staging.heimdallm.com/docs/api/meshes.md): Submit a mesh job, poll it or get a webhook, and download the GLB, with any 3D model on the marketplace.
- [Messages](https://staging.heimdallm.com/docs/api/messages.md): POST /messages — Anthropic message blocks, tools, event streams and marketplace differences.
- [Models & catalogue](https://staging.heimdallm.com/docs/api/models.md): The public catalogue: models, endpoints, providers, families, benchmarks, series, the status board and public stats.
- [Rerank](https://staging.heimdallm.com/docs/api/rerank.md): POST /rerank — score a set of documents against a query with any rerank model on the marketplace.
- [Responses](https://staging.heimdallm.com/docs/api/responses.md): POST /responses — stateless input, function tools, response objects and event streams.
- [Speech](https://staging.heimdallm.com/docs/api/speech.md): POST /audio/speech — turn text into audio with any speech model on the marketplace, billed per character.
- [Transcriptions](https://staging.heimdallm.com/docs/api/transcriptions.md): POST /audio/transcriptions — turn speech into text with any transcription model on the marketplace.
- [Video generation](https://staging.heimdallm.com/docs/api/videos.md): Submit a video job, poll it or get a webhook, and download the clip, with any video model on the marketplace.

## Guides

- [Chat apps & coding tools](https://staging.heimdallm.com/docs/guides/apps.md): Connect Open WebUI, LibreChat, Cline, Roo Code, Continue and aider to Heimdallm with the exact settings each one takes.
- [Choosing a model](https://staging.heimdallm.com/docs/guides/choosing-models.md): How to read the catalogue: prices, quantizations, measured performance, benchmarks and data policies — and how to pick between endpoints.
- [Your account](https://staging.heimdallm.com/docs/guides/dashboard.md): A tour of the account page: balances and holds, the ledger, creating and managing keys, workspace routing defaults, and activity.
- [Data & privacy](https://staging.heimdallm.com/docs/guides/data-and-privacy.md): Where your requests go, what providers declare about training and retention, how to route on it, and what Heimdallm itself records.
- [Deposits](https://staging.heimdallm.com/docs/guides/deposits.md): Funding your balance: the coins accepted, how to deposit, and what the ledger shows.
- [Working with embeddings](https://staging.heimdallm.com/docs/guides/embeddings.md): Choosing an embeddings model, batching input, and reading vectors back.
- [Troubleshooting & FAQ](https://staging.heimdallm.com/docs/guides/faq.md): Symptoms, causes and fixes for the questions buyers actually hit, from 402s to missing streams.
- [Documents & files](https://staging.heimdallm.com/docs/guides/files.md): Store a document once and name it by id in chat completions, Responses and Messages requests.
- [Gateways](https://staging.heimdallm.com/docs/guides/gateways.md): Use Heimdallm as an upstream behind Bifrost, LLMGateway or a LiteLLM proxy: the exact provider settings.
- [Images & vision](https://staging.heimdallm.com/docs/guides/images.md): Sending images to vision models: image_url content parts and the base64 data-URL requirement.
- [Organizations, workspaces & keys](https://staging.heimdallm.com/docs/guides/organizations.md): How accounts, organizations, workspaces and API keys fit together, and who can see what.
- [The playground](https://staging.heimdallm.com/docs/guides/playground.md): Use coding agents on a private machine, keep workspaces in your browser, and try chat, image and video models.
- [Production checklist](https://staging.heimdallm.com/docs/guides/production.md): Running on Heimdallm for real: retries and timeouts, fallbacks, stream discipline, cost control and monitoring.
- [Scene studio](https://staging.heimdallm.com/docs/guides/scene-studio.md): Arrange subjects and camera motion, prepare scene controls, and generate a video through the marketplace.
- [SDKs & frameworks](https://staging.heimdallm.com/docs/guides/sdks.md): Exact base URL and key settings for the OpenAI, Anthropic, Vercel AI, LiteLLM, LangChain and LlamaIndex clients.
- [Streaming](https://staging.heimdallm.com/docs/guides/streaming.md): Server-sent events on /chat/completions: frame anatomy, keep-alives, [DONE] and in-band errors.
- [Structured outputs](https://staging.heimdallm.com/docs/guides/structured-outputs.md): response_format on chat completions: json_object mode and json_schema with a worked example.
- [Tool calling](https://staging.heimdallm.com/docs/guides/tool-calling.md): Function tools on chat completions: definitions, tool_choice, and the full multi-turn loop.

## For providers

- [The provider dashboard](https://staging.heimdallm.com/docs/providers/dashboard.md): The provider surface: sources and pulls, catalogue discounts, endpoint evidence, earnings and statements, members and payout settings.
- [A worked example](https://staging.heimdallm.com/docs/providers/example.md): The reference provider end to end: eleven served entries over eight rails, the models.yaml that lists them, running it, and growing past one box.
- [Listing your models](https://staging.heimdallm.com/docs/providers/how-to-list.md): Become a provider: publish a model list document, pass health probes, and get paid for the traffic you serve.
- [Operating a source](https://staging.heimdallm.com/docs/providers/operating.md): Running a listing in production: pull outcomes and what each does, health and lifecycle, capacity reporting, dialects, and the common findings.
- [Pricing & demand](https://staging.heimdallm.com/docs/providers/pricing.md): How price steers routing, how to be found and trusted, and the economics of serving on the marketplace.

## Optional

These are rendered pages with no markdown form. The catalogue behind them is also an API, which is
the better way to read it.

- [Model catalogue](https://staging.heimdallm.com/models): every model, the providers serving it, their prices and their measured performance.
- [Provider status](https://staging.heimdallm.com/status): live health, capacity and uptime, measured from probe traffic.
- [Activity](https://staging.heimdallm.com/activity): what the marketplace served, and models ordered by measured usage.