Ainfera — Platform API

The Inference of AI Agents. L2 routing runtime — /v1/inference, workflows, audit chain.

This repository is the public Ainfera platform: FastAPI service, L1–L5 layers (Identity, Routing, Settlement, Audit, Orchestration), provider adapters, and Alembic migrations. Deployed at api.ainfera.ai.

What Ainfera does

Constrained routing across 50+ models. One Agent Card per agent. Every call cryptographically audited. pip install ainfera.

L1–L5

• L1 Identity — Ed25519 keypair + JWS RFC 7515 AgentCards
• L2 Routing — ProviderAdapter dispatch (Anthropic, OpenAI, Together)
• L3 Settlement — Internal prepaid ledger with atomic debit + per-call spend cap
• L4 Audit — Hash-chained, HMAC-signed append-only event log
• L5 Orchestration — FastAPI REST surface, OpenAPI, MCP server

Quickstart (local)

uv sync
uv run alembic upgrade head
uv run uvicorn ainfera_api.main:app --reload

Quickstart (consume)

curl -X POST https://api.ainfera.ai/v1/agents/signup \
  -H "Content-Type: application/json" \
  -d '{"agent_handle": "my-bot"}'

Repository identity

• This repo (ainfera-ai/api): the platform external + internal agents consume via HTTPS or the ainfera SDK.
• Internal Ainfera company OS (5 agents: Varda, Namo, Aule, Yavanna, Tulkas) lives in ainfera-ai/ainfera-os — consumed via pip install ainfera, identical to external customers.
• Client SDK: ainfera-ai/sdk
• Public specs: ainfera-ai/specs
• Audit verifier: ainfera-ai/verify
• MCP server: ainfera-ai/mcp-server

See docs/ADR-010 for the split rationale.

License

Proprietary today · Apache 2.0 when public (Sprint v1.8, per D-052).