Fire is the inference engine underneath Modus Promethean — a single, model-agnostic API that routes chat, image, and speech generation across 13 providers, logs every call it will ever make, and never lets an agent touch a raw credential.
Most "AI gateways" are a thin proxy in front of one provider's SDK. Fire is the pipeline every app in the Modus Promethean stack — Saga, FireScout, ImageLab, PornArt — actually runs production traffic through. What's sold here is exactly what's dogfooded internally.
One request shape for 30+ models across Anthropic, OpenAI, Google, Mistral, DeepSeek, and more.
Swap species_name, not your integration.
Every chat, image, and speech call writes to prompt_logs inside a finally block —
it fires on success or failure, with no parameter to suppress it.
Compose DAGs of model calls with parallel branches and real human-in-the-loop gates. Queued, resumable, billed per step — not a single opaque black box.
Need to reach a third-party service? Fire injects the credential server-side and forwards the call — your agent's token is the badge, it never sees the key.
Fire doesn't pick a lane. It routes each request to the model best suited for the job, and meters the same way regardless of what's on the other end of the wire.
Structured extraction, sentiment, and reasoning over filings, tickers, and market data — routed to reasoning-tier models when the analysis actually needs it.
Feed papers, datasets, and molecular data through long-context models built for reasoning across hundreds of pages, not chatting about them.
Route to coding-tuned models for generation and review, with every prompt and diff logged — useful the day someone asks what the agent actually did.
Get a key from the portal, then you're one request from a response — with the price you were actually charged returned inline, every time.
Every call returns meta.price.usd — the actual amount charged, always,
even when it's zero. Fire's own provider cost is never mixed into that number.
The default for new accounts. Text is billed off a daily-synced published price list; images are a flat per-image rate. What you see in the docs is what you pay — it doesn't move between calls.
For infra partners who want real provider cost plus a flat, disclosed markup instead of the retail list — computed fresh on every single call.
A model isn't sellable until it passes a live checklist: a real call succeeds right now, usage is reported, cost is verified, a price is published. A weekly audit re-runs the same checklist — a model that dies upstream is pulled automatically, not left silently broken.
Fire is a partial reference implementation of the AIGCSEP protocol (covenant.to) — a three-plane architecture separating governance, telemetry, and data, with emergency-stop hooks defined and authenticated ahead of the day they're needed.
A token with no resolvable billing tier is refused before any provider is ever called — a misconfigured account can't generate spend with nowhere to send the bill.
Free to sign up — you only pay when you top up. No card required to get your first key.