Fermi

Ask a question like “What will the price of Bitcoin be in 6 months?” and Fermi returns a probability distribution — family, parameters, coverage interval, point estimate — with reasoning and assumptions. Works for market sizing, forecasting, risk assessment, and any question where you need a calibrated probabilistic answer.

Getting Started

Fermi is a REST + MCP API. The fastest way to explore:

Browse the interactive API docs (Swagger UI) — try requests live, see schemas, and use the Authorize button to set your API key.
Create an account — curl -X POST https://fermi.krobar.ai/api/v1/accounts — returns an API key and 3 starter credits.
Make your first estimate:

curl -X POST https://fermi.krobar.ai/api/v1/estimates \
  -H "X-API-Key: <your api_key from step 2>" \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{"question": "How many coffee shops are in SF?",
       "requested_tier_id": 1, "unit": "count",
       "target_horizon": "current",
       "disclaimer_acknowledged": true}'

Use cases

Market sizing — “How many coffee shops are in San Francisco?”
Forecasting — “What will AAPL close at next Friday?”
Risk assessment — “What is the probability of a recession in the next 12 months?”
General estimation — any question where a calibrated probability distribution is useful

What you get back

Every estimate includes:

Estimation interval with coverage probability (e.g., 90% CI)
Distribution family and parameters (normal, log-normal, beta, etc.)
Point estimate and confidence level
Reasoning summary explaining the estimate
Assumptions made during estimation

API Documentation

Interactive API docs (Swagger UI) — includes Authorize button for API key
API reference (ReDoc)
OpenAPI schema (JSON)

Quick Start

If you are an AI agent, here is everything you need to make your first estimate. The full self-contained flow is below; the linked files are for deeper reference.

Auth rules (read these once, save yourself a retry):

Pass your API key as the X-API-Key header. Do NOT put it in Authorization: Bearer — that header is reserved for JWTs and will return 401 Invalid bearer token if you try it with a raw API key.
Every POST /api/v1/estimates call requires an Idempotency-Key: <uuid> header. Generate a fresh UUIDv4 per call. Omitting it returns 400.
The request body must include "disclaimer_acknowledged": true.

Full cold-start flow (copy-pasteable):

# 1. Sign up (returns {account_id, api_key, created_at}; 3 starter credits)
curl -X POST https://fermi.krobar.ai/api/v1/accounts

# 2. Make an estimate (replace <api_key> with what you got in step 1)
curl -X POST https://fermi.krobar.ai/api/v1/estimates \
  -H "X-API-Key: <api_key>" \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{
    "question": "What will the price of Bitcoin be in 6 months?",
    "requested_tier_id": 1,
    "unit": "USD",
    "target_horizon": "6 months",
    "disclaimer_acknowledged": true
  }'

The estimate response contains estimation_interval {low, high}, distribution_family, distribution_parameters, point_estimate, reasoning_summary, assumptions, and credits_charged.

Agent Discovery Files

/llms.txt — concise overview in llmstxt.org format
/agents.md — longer quickstart with Python examples and response-schema walkthrough
/openapi.json — machine-readable OpenAPI schema
/.well-known/mcp.json — MCP endpoint descriptor
/docs — Swagger UI (interactive)

MCP Integration

Fermi supports the Model Context Protocol:

HTTP adapter: GET /mcp/tools → POST /mcp/tools/estimate
SSE transport: /mcp/sse/ for native MCP clients