Experiential

// Overview

Experiential is an open-source gateway and router for agent workflows. It does three things: it puts hosted, BYOK, local and custom models behind one OpenAI-compatible API; it controls which users and agents may use which models, for which use cases, and how much they can spend; and it turns production traffic into a custom router — or a custom model — optimised for quality, speed and cost.

The local gateway is a pip install and a single command. On first run a setup wizard walks through provider connections, model and reasoning-effort selection, then shows defaults for the public alias, the identity and a command budget before printing a one-time key. From there the gateway serves the usual /v1/chat/completions surface on loopback, backed by a compiled native data plane, and a Python helper loads a fitted project router as an ordinary OpenAI client.

The optimisation path starts from OpenTelemetry traces of your existing agent: exp build walks you through providers, models and budget and constructs a simulation from those traces, against which a router is optimised; exp optimize model then fine-tunes an open-source model you own. A hosted gateway serving the same OpenAI-compatible and Anthropic Messages APIs is available if you would rather not run one, and the repository ships copy-paste setup prompts you can hand to a coding agent. Anonymous aggregate product telemetry is on by default and can be turned off with exp config telemetry disable.

// What it does

• One OpenAI-compatible API over hosted, BYOK, local and custom models, with the hosted gateway also serving the Anthropic Messages API
• Per-identity governance: which users and agents can reach which models, for which use cases, and under what spend budget
• Router fitting from your own OpenTelemetry traces — exp build constructs a simulation and optimises a router against it
• exp optimize model fine-tunes an open-source model you own from the traffic the router collected
• Local single-command gateway with a first-run setup wizard, a compiled native data plane, and a one-time issued key
• exp.load_router() returns a fitted project router as an official OpenAI client backed by its own private gateway

// Getting started

Install the package and run exp; the first-run wizard handles providers, the public alias, the identity and a command budget, then prints a one-time key.

pip install experiential
exp

export EXP_GATEWAY_KEY=...
curl http://127.0.0.1:8000/v1/chat/completions \
  -H "Authorization: Bearer $EXP_GATEWAY_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"model":"opus-5","messages":[{"role":"user","content":"Help me"}]}'

exp build support-agent

import exp

with exp.load_router("my-project") as client:
    response = client.chat.completions.create(
        model="my-project",
        messages=[{"role": "user", "content": "hello"}],
    )

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Give a team one endpoint and one key format across several model providers, without rewriting each agent for each vendor
• Cap what an individual agent or user can spend, and restrict which models it may reach, before the bill arrives
• Repoint existing coding agents (Claude Code, Cursor, Codex, Aider) at a gateway you control so their traffic is logged and governed
• Turn recorded production traces into a cheaper routing policy, or into a fine-tuned open-source model you own

// How Experiential compares

Experiential alongside other open-source gateways & routing tools AI/TLDR tracks, ranked by GitHub stars.