OpenFugu

Transparent multi-model routing for your own model pool.

An OpenAI-compatible gateway that orchestrates local and remote models with configurable policies. See every route, cost, latency, retry, verifier, fallback, and baseline comparison before the final answer is returned.

POSThttps://api.openfugu.dev/v1/chat/completions
route report

openfugu/default

schema preview
cost
runtime field
latency
runtime field
baseline
configured field
stepmodelcostlatencyresult
classifypolicy enginerecordedrecordedtask category
selectworker registryrecordedrecordedcandidate workers
verifyoptional verifierrecordedrecordedcheck decision
fallbackfallback chainrecordedrecordedif triggered
Gateway surface

A model orchestration layer, delivered as one endpoint.

OpenFugu sits in front of the model pool you already use. It does not hide dispatch behind a black box: policy, route choice, worker role, verifier result, fallback attempt, cost, latency, and fixed-model comparison become inspectable system output.

route report

openfugu/default

schema preview
cost
runtime field
latency
runtime field
baseline
configured field
stepmodelcostlatencyresult
classifypolicy enginerecordedrecordedtask category
selectworker registryrecordedrecordedcandidate workers
verifyoptional verifierrecordedrecordedcheck decision
fallbackfallback chainrecordedrecordedif triggered

Local runtimes

  • llama.cpp
  • Ollama
  • LM Studio

Remote APIs

  • OpenAI
  • Claude
  • Gemini

Open models

  • DeepSeek
  • Qwen
  • Llama / Gemma

Private endpoints

  • vLLM
  • Internal APIs
  • Team services
eval replay

Turn each route into a replayable comparison sample.

arearecorduse
input profiletask constraintsCapture task type, context size, privacy constraints, quality risk, and expected output before judging the route.
route reasonpolicy decisionStore matched policies, candidate workers, selected worker, excluded workers, and the routing explanation for audit.
baseline runfixed worker outputRun the same input through one fixed worker so the team can inspect answer differences, failure points, and human preference.
failure loopretry reasonsFeed verifier rejections, fallback triggers, timeouts, and low-confidence cases back into the next eval set and future router training.
Trace signals

What the route makes visible.

01

OpenAI-compatible

Drop into tools that already speak the OpenAI API without changing callers.

02

Transparent trace

See dispatch, workers, prompts, retries, verifier output, and intermediate results.

03

Cost and latency visible

Track estimated cost, token usage, timings, retries, and provider failures.

04

Verifier and fallback

Check risky answers and recover from timeouts, low confidence, or bad outputs.

05

Baseline comparison

Compare the route against a fixed model to measure real improvement.

Policy layer

Compose simple policies now. Train a conductor when the data exists.

The first OpenFugu beta should be useful before training: explicit policies, traceable plans, and eval replay from real user tasks.

Request beta access

cheapest-good-enough

Start with the lowest-cost worker that meets the quality bar.

verify-before-return

Let a stronger model check the answer before it reaches the caller.

fallback-chain

Retry with the next viable worker when confidence or reliability drops.

fixed-model-baseline

Measure orchestration against one fixed worker, not vibes.

Validation path

Validate demand first, expand capability next.

OpenFugu validates whether transparent routing solves real problems before expanding prototypes, evals, and model support.

01

Demand validation

Confirm who truly needs transparent multi-model routing, which tools they use, and what model pools they operate.

02

Transparent routing prototype

Show route traces, cost records, latency records, failed retries, and fixed-model comparisons.

03

Beta trial

Invite users with concrete model pools and real tasks, then collect replayable eval samples.

04

Capability iteration

Use real feedback to prioritize strategies, worker/provider support, and deployment shapes.

Join waitlist

Help decide the first OpenFugu beta.

Tell us which tools you use, which models belong in your pool, where fixed models fail, and whether you need a local client, Docker gateway, team service, or cloud option.

1POST /v1/chat/completions
2{
3 "model": "openfugu/default",
4 "policy": "quality_balanced",
5 "cost_budget": "<configured_limit>",
6 "verifier": true,
7 "baseline": "<fixed_worker>"
8}