Inferix documentation
Inferix control plane docs: observe, route, detect drift, and retrain for AI inference and agents.
Inferix is the control plane for AI inference and agents. Observe every call, route by policy, detect quality drift, and retrain when signal says so. It sits next to your gateway and traces — not instead of them.
The loop
Platform teams run one control plane in front of owned models and providers. Traffic flows through the same path every time:
- LensAI — latency, cost, errors, volume
- TraceForge — spans across tools, models, and agent steps
- RouteIQ — policy: which model handles this call
- DriftWatch — quality slips before users complain
- FineForge — fine-tune, promote, or roll back
Get started
Quick start
lensai-integration: make build / up / smoke → Grafana :3000.
Local workspace
Multi-repo workspace: ./run-local.sh → console :3003 + full stack.
Learn: beginner → owned models
What an LLM is, train a tiny GPT, then register it on Inferix.
Install & self-host
Real Compose and Helm from LensAI + streaming — no fake image.
Status & maturity
Active vs Scaffold vs Design target for every product.
Core concepts
Control plane, call, span, policy, drift, promote — glossary.
Architecture
Shipped LensAI pipeline vs design-target control plane.
Operate
Configuration
Full inferix.yaml reference for models, routing, observe, and drift.
Observability
LensAI metrics and TraceForge spans together.
Routing policies
RouteIQ defaults, cheap SLMs, strong paths, tenant rules.
Drift & retrain
DriftWatch signals and FineForge promote / rollback.
API reference
Chat completions, health, models, traces, policies, alerts.
SDKs & clients
Python, JS, Go patterns and OpenTelemetry for TraceForge.
Products
LensAI
See latency, cost, errors, and volume across models and agents.
TraceForge
Follow one request across tools, models, and agent steps.
RouteIQ
Send each call to the right owned model or provider by policy.
DriftWatch
Alert when quality slips — before users complain.
FineForge
Turn drift into a fine-tune job you can ship or roll back.
Guides
Practical agent and operator playbooks — from support CX and incident SRE to voice, API healing, and owned SLMs.
Learn: beginner → owned models
What an LLM is, train a tiny GPT, then register it on Inferix.
All guides
Index of every operator and agent guide.
Platform operators
Run the control plane for ML and platform teams.
Owned models
Register SLMs, cheap vs strong RouteIQ paths, FineForge.
Who this is for
Platform and ML ops teams who own inference cost, quality, and agent reliability. If you already have a gateway or an APM, Inferix sits next to them and closes the observe → route → drift → retrain loop.