Which base URL and auth field do I use?

OpenAI SDKs use https://api.mindshub.ai/v1 with api_key. Anthropic SDKs use https://api.mindshub.ai without /v1 and must use auth_token — not api_key — so the client sends the Authorization: Bearer header.

How does billing work?

There is no subscription. The MindsHub Air model, alias mindshub_air, is free within fair use, with daily limits the API enforces. Other usage draws from a prepaid organization balance at the rates on the pricing page.

Does conversation chaining work?

No. Keep conversation history client-side and send the full history on each turn. Prompt caching makes repeated prefixes cheaper on most models; previous_response_id and store are accepted but not honored.

How do aliases and availability work?

Every model has a fixed alias that names one exact version, such as gpt-6-astra, and never moves. Most lines also have a rotating alias, such as gpt, sonnet, or kimi, that resolves to the most recent version of that line MindsHub serves, so it takes upgrades without a code change. Either string goes in the model field. GET /v1/models shows the live catalog and what your organization can use.

Can I bring my own provider key?

Not today. Bring-your-own-key is planned, not available. The current service uses one MindsHub key and one organization wallet.

What is the MindsHub Foundry?

Selected teams work directly with the product team, test new features, and provide regular feedback. Members receive product support, early access, and a 100% match on their first three months of inference-credit purchases, up to $5,000. The program is open worldwide, places are limited, and selection is not guaranteed.

How does this relate to MindsHub Cowork?

Cowork is a ready-made workspace that uses the same model catalog. Unified Inference is the API for your own software. They share one account and balance.

Unified Inference

Call leading models through one API.

Use OpenAI or Anthropic SDKs to call the same model catalog with one key and one prepaid balance.

Get an API key Start with the docs

OpenAI Chat Completions
Anthropic Messages
Pay as you go
Free MindsHub Air model

Switch an existing client

Change the base URL, not your client.

Point your code at api.mindshub.ai and use a short model alias. Copy the command on the right to test the endpoint.

OpenAI SDKs use https://api.mindshub.ai/v1 with api_key.
Anthropic SDKs use https://api.mindshub.ai without /v1 and require auth_token, so the client sends Authorization: Bearer.

Which API should I use? →

curl https://api.mindshub.ai/v1/chat/completions \
  -H "Authorization: Bearer $MINDSHUB_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sonnet",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Short aliases Switch models by changing one string
Three request formats Use the client you already have
Tools and streaming Keep each protocol’s event format
Usage billing One prepaid organization balance

Free tier

The MindsHub Air model is free for your agent.

Point OpenClaw, Hermes, Claude Code, Codex or your own code at api.mindshub.ai with one key and use Air at no charge.

"model": "mindshub_air"

The free tier is for people using agents. Daily rate limits apply, and accounts that run automated bulk traffic through it can be suspended. If a limit stops you and you think that's a mistake, contact us.

Get an API key Read the fair-use policy →

Invitation · limited places

Join the MindsHub Foundry

Test the API with direct access to the product team

We’re looking for teams willing to use the API in production and tell us where it falls short.

Members receive · spend matching

100% match up to $5,000

A 100% match on your first 3 months of inference-credit purchases, up to $5,000 — matched against the credits you buy, not a grant.

Members receive

Direct product and technical support
Early access to new features and capabilities

Who we’re looking for

Teams already building or operating AI products, workflows, or services that:

Plan to use MindsHub Inference during the program
Can share specific feedback from that work
Can join regular calls with our product team

Applications are reviewed on a rolling basis and take about 3–5 minutes to complete. Applying does not guarantee acceptance.

The application form needs JavaScript, so the buttons above won’t open it. Use the contact form and mention the Foundry — we’ll send you the application.

Choose your API

Use the request format you already have.

All three formats reach the same catalog and the same bill. Start with Chat Completions for new work; use Messages for Anthropic SDKs and Claude Code.

OpenAI Production

Chat Completions

POST /v1/chat/completions

Use the OpenAI SDK with api_key and a base URL ending in /v1.

Open the Chat Completions guide →

Anthropic Production

Messages

POST /v1/messages

Use Anthropic SDKs or Claude Code with auth_token and the base URL without /v1.

Open the Messages guide →

OpenAI Mid-upgrade

Responses

POST /v1/responses

Input is accepted today, but Responses-shaped output and SDK helpers are still upgrading. Use Chat Completions for production.

Read the current Responses status →

Model aliases

Name a model by alias: follow the line, or pin a version.

Every model has a fixed alias that names one exact version and never moves. Most lines also have a rotating alias that always resolves to the most recent version of that line we serve, so you take upgrades without a code change. Put either string in model; GET /v1/models shows what your organization can use.

Maker	Model	Rotating alias	Fixed alias
MindsHub	MindsHub Air Free tier	`mindshub_air`	—
Anthropic	Claude Fable 5.1	`fable`	`fable-5-1`
	Claude Opus 5	`opus`	`opus-5`
	Claude Fable 5	—	`fable-5`
	Claude Sonnet 5	`sonnet`	`sonnet-5`
	Claude Haiku 4.5	`haiku`	`haiku-4-5`
OpenAI	GPT-6 Astra	`gpt`	`gpt-6-astra`
	GPT 5.6 Sol	—	`gpt-5-6-sol`
	GPT 5.6 Terra	`gpt-terra`	`gpt-terra-5-6`
	GPT 5.6 Luna	`gpt-luna`	`gpt-luna-5-6`
	GPT 5.3 Codex	`gpt-codex`	`gpt-codex-5-3`
	GPT 5.4 Mini	`gpt-mini`	`gpt-mini-5-4`
	GPT 5.4 Nano	`gpt-nano`	`gpt-nano-5-4`
Google	Gemini 3.8 Flash	`gemini-flash`	`gemini-flash-3-8`
	Gemini 3.7 Flash	—	`gemini-flash-3-7`
	Gemini 3.6 Flash	—	`gemini-flash-3-6`
	Gemini 3.5 Flash	—	`gemini-flash-3-5`
	Gemini 3.1 Pro Preview	`gemini`	`gemini-3-1-pro`
	Gemini 3.1 Flash-Lite	`gemini-flash-lite`	`gemini-flash-lite-3-1`
	Gemini 3 Flash Preview	—	`gemini-flash-3`
Moonshot AI via Fireworks AI	Kimi K3	`kimi`	`kimi-k3`
DeepSeek via Fireworks AI	DeepSeek V4 Pro	`deepseek`	`deepseek-v4-pro`
	DeepSeek V4.1 Flash	`deepseek-v4-flash`	`deepseek-v4-1-flash`
	DeepSeek V4 Flash	—	`deepseek-v4-flash-0731`
Alibaba via Fireworks AI	Qwen3.8-2.4T-A95B	`qwen`	`qwen-3-8-a95b`
Zhipu via Fireworks AI	GLM 5.3	`glm`	`glm-5-3`
	GLM 5.3 Flash	`glm-5-3-flash`	`glm-flash-5-3`
	GLM 5.2	—	`glm-5-2`
Meta	Muse Spark 1.3	`muse-spark`	`muse-spark-1-3`
	Muse Spark 1.2	—	`muse-spark-1-2`
	Muse Spark 1.1	—	`muse-spark-1-1`
SpaceXAI	Grok 4.6	`grok`	`grok-4-6`
SpaceXAI	Grok 4.5	—	`grok-4-5`
Cerebras	MindsHub Blaze	`mindshub_blaze`	—
Fireworks AI	GPT-OSS 120B (Fireworks)	—	`gpt-oss-fireworks`

See prices on the rate card Compare them on quality, speed and cost → Model aliases + GET /v1/models →

Pay as you go

One balance for the whole catalog.

The MindsHub Air model, alias mindshub_air, is the free tier within fair use. Every other model, embeddings, web search, and priced cache writes draw from one prepaid organization wallet.

Input, output, cached input, and cache writes are metered separately. Cached input is roughly a tenth of the ordinary input rate. Daily limits on the free tier are enforced by the API; your entitlements response is authoritative.

Read the billing docs See the rate card →

Conversation history / repeated prefix Metered

First request · write System prompt + history input + cache write

Request 02 · read Repeated prefix ~1/10 input rate

Request 03 · read Repeated prefix ~1/10 input rate

Request 04 · read Repeated prefix ~1/10 input rate

Caching is automatic on most of the catalog. Claude-family models use cache_control breakpoints on Messages.

Parameter adaptation

Handle common parameter differences.

Supported parameters pass through. Unsupported ones are dropped and out-of-range values are clamped. Response headers name every change.

POST /v1/chat/completions        model: gemini-flash
  top_k · max_tokens · reasoning_effort

← 200 OK
  X-MindsHub-Dropped-Params: top_k
  X-MindsHub-Clamped-Params: max_tokens=requested>applied

Nothing changed? No adaptation headers. Provider value restrictions can still return a 400.

Streaming

Keep your existing stream parser.

Chat Completions and Messages keep their own event, completion, and error formats.

Compare API behavior →

Tools + images

Send tools, images, and search requests.

Function tools, image input, and built-in web search are supported where the selected model and request format support them.

See working examples →

Usage

Review usage by model.

API and agent requests appear in the same organization usage summary, grouped by model.

Usage and wallet docs →

One engine

All formats use the same catalog and balance.

Every request format reaches the same models, the same wallet, and the same usage summary.

Compare all three APIs →

POST /v1/chat/completions Production
POST /v1/messages Production
POST /v1/responses Mid-upgrade
GET /v1/models Live
POST /v1/embeddings Live
Usage summary Live

Want a ready-made workspace?

Use the same models in Cowork.

Describe a project, connect its data, and get back documents, dashboards, or apps. Cowork uses the same catalog and balance, with no client to build.

Explore Cowork Start in your browser →

MindsHub Cowork — a briefed project running in the workspace, with its finished artifacts listed alongside.

FAQ

Questions before you switch.

Which API should I use?: Use Chat Completions for new integrations and Messages for Anthropic SDKs or Claude Code. Responses is still being upgraded, so use Chat Completions for production.
Which base URL and auth field do I use?: OpenAI SDKs use https://api.mindshub.ai/v1 with api_key. Anthropic SDKs use https://api.mindshub.ai without /v1 and must use auth_token — not api_key — so the client sends the Authorization: Bearer header.
How does billing work?: There is no subscription. The MindsHub Air model, alias mindshub_air, is free within fair use, with daily limits the API enforces. Other usage draws from a prepaid organization balance at the rates on the pricing page.
Does conversation chaining work?: No. Keep conversation history client-side and send the full history on each turn. Prompt caching makes repeated prefixes cheaper on most models; previous_response_id and store are accepted but not honored.
How do aliases and availability work?: Every model has a fixed alias that names one exact version, such as gpt-6-astra, and never moves. Most lines also have a rotating alias, such as gpt, sonnet, or kimi, that resolves to the most recent version of that line MindsHub serves, so it takes upgrades without a code change. Either string goes in the model field. GET /v1/models shows the live catalog and what your organization can use.
Can I bring my own provider key?: Not today. Bring-your-own-key is planned, not available. The current service uses one MindsHub key and one organization wallet.
What is the MindsHub Foundry?: Selected teams work directly with the product team, test new features, and provide regular feedback. Members receive product support, early access, and a 100% match on their first three months of inference-credit purchases, up to $5,000. The program is open worldwide, places are limited, and selection is not guaranteed.
How does this relate to MindsHub Cowork?: Cowork is a ready-made workspace that uses the same model catalog. Unified Inference is the API for your own software. They share one account and balance.