Private LLM inference, chat and API for financial firms

Private chat for your whole firm and an OpenAI-compatible API for your developers. Open-weight models handle the work by default. Hybrid routing to Claude, OpenAI or Amazon Bedrock is off until your firm turns it on.

Ways to run models

Choose how each workload runs, behind one API

Every option uses the same keys, the same OpenAI-compatible endpoint and the same controls, so you can move a workload between them without rewriting it.

Start today

Pooled capacity

Open-weight models on pooled capacity, billed per token. Sign up under our fair use policy and make your first call in minutes.

Hybrid routing

Off by default. When an administrator turns it on, requests to Claude, OpenAI or Amazon Bedrock models leave our network through the same API, keys and controls.

Ask about hybrid routing Half price

Batch

Send large jobs that do not need an instant answer and pay half the normal token rate, with an OpenAI-compatible Batch API.

Read the Batch API docs Dedicated

Dedicated endpoints

Serve catalogue models or your own weights on dedicated GPUs with vLLM or SGLang, callable through the same API.

Control Plane

See model pricing

Model aliases

Aliases that keep working when a model is unavailable

Aliases such as fast, reasoning and coding point at the best model for the job. If that model is unavailable, the alias routes to another, so your applications keep running. Live service status is always at status.privatemind.com.

View service status

fast: The most efficient model available, for everyday requests where speed matters.
reasoning: The strongest model for hard problems and long-running agents.
coding: Tuned for agentic coding with a long context window.
default: A balanced choice with vision, thinking and tools.

Chat

A chat assistant your compliance team can approve

Choose from every model your firm has enabled, attach files and connectors, and keep work organised. Administrators decide what is switched on.

Projects: Group related conversations and pin the sources they need, so every new chat in a project starts with the right context.
Sources: Upload PDFs, documents and spreadsheets for retrieval, and share them with a colleague or a whole team.
MCP connectors: Connect the MCP servers and APIs your firm approves, including Microsoft 365 and AtlasFeed market data.
Skills: Turn on packaged procedures such as building a cash-flow model, or write your own in the open Agent Skills format.
Memory: When someone opts in, the assistant remembers durable facts from their own messages, private to them.
Cloud agents: Send a long task to run in the background, or schedule a prompt to run every morning and report back.
Web search: Turn on web search for questions that need current information, where your organisation allows it.
Guardrails: Chats run with guardrails that flag unsafe prompts and responses, in line with our fair use policy, and administrators decide which models and tools people can use.

PrivateMind

Microsoft 365
SharePoint
Teams
Slack
Notion
Confluence
Jira
GitHub
Datadog
Snowflake
AtlasFeed market data
Your databases
MCP servers

Built-in connectors for Microsoft 365, plus any MCP server your firm approves. Permissions carry through to every answer.

Tenant isolation

Every client gets its own database

PrivateMind is multi-tenant, but your data does not share a database with anyone else's. Each client is kept in a separate database, with teams and projects on every plan. Enterprise contracts add the compliance API, audit log export and SLAs.

API

A private LLM inference API that is OpenAI-compatible

Point the official OpenAI SDKs, LangChain, LlamaIndex or the Vercel AI SDK at our base URL. Chat, tools, structured output, embeddings, rerank, vision, audio, images, files and batch are all supported.

Read the quickstart OpenAI SDK guide

cURL

curl https://api.privatemind.com/v1/chat/completions \
  -H "Authorization: Bearer $PMIND_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "reasoning",
    "messages": [{"role": "user", "content": "Draft the morning risk summary"}]
  }'

Questions

Platform questions

Something not covered here? Talk to our team or read the documentation.

What is the PrivateMind Platform?

Private LLM inference for financial firms: a chat assistant for your whole firm and an OpenAI-compatible API for your developers. Open-weight models handle the work by default, and hybrid routing to Claude, OpenAI or Amazon Bedrock is off until your firm turns it on.

Does the API work with OpenAI SDKs?

Yes. Point the official OpenAI SDKs, LangChain, LlamaIndex or the Vercel AI SDK at our base URL. Chat, tools, structured output, embeddings, rerank, vision, audio, images, files and batch are all supported.

Is our data kept separate from other clients?

Yes. PrivateMind is multi-tenant, but each client is kept in a separate database, so your data does not share a database with anyone else’s. Enterprise contracts add the compliance API, audit log export and SLAs.

What happens if a model becomes unavailable?

Model aliases such as fast, reasoning and coding point at the best model for the job. If that model is unavailable, the alias routes to another, so your applications keep running.

Put private AI in front of your whole firm

Sign up to try chat and the API today, or talk to us about rolling PrivateMind out across your teams with an enterprise SLA.