Welcome to Mithril

One API. Verified researchers. OpenAI-compatible — tool calls and streaming included.

Mithril is an OpenAI-compatible inference API. Point an existing OpenAI SDK at https://api.mithril.fund/v1 and call security-research-tuned models behind guardrails, a firewall and audit receipts. The verified free allowance stays; add prepaid credits only as you need them.

Why Mithril

• OpenAI-compatible by design

  Every request and response is in the OpenAI JSON shape. Point base_url at https://api.mithril.fund/v1 and streaming, tool calling and reasoning keep working. Anthropic Messages (POST /v1/messages) is not served (405) — Claude Code-class clients need a translating proxy.

• Admission for researchers

  Red-team models run only behind card-based identity verification, AUP consent and an approved scope; blue-team models need a sign-in. Guardrails run on every request either way.

• Prepaid, no markup on the meter

  1 credit = $1. Input $0.9/M and output $4.5/M tokens, settled from the provider receipt. Automatic recharge is opt-in, off by default; with no balance you are simply on the free allowance.

• Zero data retention

  Prompts and outputs are processed in memory; no inference logs. What remains is verification status, evidence references and audit receipts.

Where to start

• Getting started (console)

  Make your first call in five minutes.

• Compatibility

  OpenAI-compatible setup for Codex CLI, Cursor, Cline, Aider, hermes. Claude Code needs a translating proxy — Anthropic Messages is not served yet.

• Advanced

  Streaming, tool calling, structured outputs, reasoning.

• API reference

  Every endpoint, error, limit and charge.

• Security services (CTEM / DAST / SAST / SPECT / VM / GRC / IR / DR)

  The public catalog at GET /v1/security/services. DAST (gated) probes running applications and reports runtime findings with request/response evidence to the VM ledger.

• Graph & ontology database

  Kotobase: content-addressed graph state and Datalog.