Ainfera Routing

The Inference of AI Agents. Agent-native inference routing — constrained optimization across 50+ models, not a billing gateway.

Ainfera Routing is the control plane for production agent workloads: pick the highest-quality model that satisfies hard budget and latency limits, dispatch, capture outcomes, and compound routing intelligence from real traffic.

Why this exists

Gateways like LiteLLM, OpenRouter, and Portkey optimize single calls for human teams. Ainfera Routing targets autonomous agents: hard caps, workflow-aware traces, identity-bound receipts, and an outcome loop that only compounds when traffic flows through Ainfera.

Quick links

Resource	URL
Methodology (Notion)	Ainfera Routing — Methodology v1.0
API	api.ainfera.ai
Python SDK	ainfera-ai/sdk
Specs	ainfera-ai/specs
MCP	mcp.ainfera.ai

Starter routing templates

Copy a template into your tenant routing policy or agent config:

Template	Goal	File
Latency-first	Minimize p95 under quality floor	templates/latency-first.yaml
Cost-first	Minimize cost under quality floor	templates/cost-first.yaml
Compliance-first	Strict veto + no fallback	templates/compliance-first.yaml

Policy shape is documented in schema/routing-policy.schema.json.

Minimal curl

export AINFERA_API_KEY="ai_infera_..."

curl -sS https://api.ainfera.ai/v1/inference \
  -H "Authorization: Bearer $AINFERA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ainfera-inference",
    "messages": [{"role": "user", "content": "Route this agent call"}],
    "max_tokens": 256
  }'

Use model: "ainfera-inference" to let Ainfera route across the live catalog. (ainfera-mithril / ainfera-auto are legacy aliases and still resolve identically.) See docs for Agent Cards, workflows, and audit verification.

Production E2E smoke

Full L1–L5 verification against live api.ainfera.ai (~3 min):

git clone https://github.com/ainfera-ai/routing
cd routing
./scripts/ainfera-e2e.sh          # human-readable
MODE=json ./scripts/ainfera-e2e.sh   # agent-parseable

Spec: scripts/AINFERA-E2E.md

Outcome capture (methodology §16): docs/outcome-capture.md

Repo map (ainfera-ai org)

Repo	Role
routing (this repo)	Public methodology, templates, onboarding
api	L2 routing runtime + /v1/inference
sdk	Python client
specs	ATS, AAMC, API contracts (CC-BY 4.0)
ainfera-mcp	MCP reference adapter
Framework adapters	ainfera-mcp, ainfera-hermes, ainfera-openclaw, ainfera-letta, ainfera-langgraph, ainfera-langchain, ainfera-llamaindex, ainfera-crewai, ainfera-google-adk

License

Apache 2.0 — see LICENSE.