Experiential

Overview

Experiential is an open-source AI gateway that sits between your agents and inference providers. Instead of scattering API keys and base URLs across Claude Code, Cursor, Codex, Aider, and other tools, you run one local gateway (pip install experiential; exp) or point at the hosted platform, and route all traffic through a single OpenAI-compatible endpoint. The gateway adds zero markup on inference pass-through and provides a unified control plane for spend, access, and optimization. With 12.7K+ GitHub stars and 264 forks, it is gaining rapid traction as an agent infrastructure layer.

Key Capabilities

Experiential’s core value is turning production traffic into intelligence. The exp build command ingests OpenTelemetry traces from your existing agent workflows and builds a project-specific router that balances quality, speed, and cost across providers. The exp capture command collects traces directly from Claude Code and Codex desktop apps using your existing subscriptions — no code changes required. Once you have enough traffic data, exp optimize model fine-tunes an open-source model on your traces, giving you a specialized model you own.

The gateway enforces granular access controls: define which users and agents can use which models, for which use cases, and how much they can spend. Budgets are set per identity with configurable limits (default $50 per command). Provider keys for OpenAI, Anthropic, Gemini, Azure, Bedrock, Fireworks, and OpenRouter are managed centrally through the BYOK connection flow.

Use Cases

Development teams use Experiential to consolidate agent infrastructure. Rather than configuring each coding agent with its own API keys and provider settings, the gateway centralizes provider connections and enforces consistent spend limits. Teams building production agent systems use the traffic optimization pipeline to reduce inference costs by routing to cheaper models where quality is maintained. Organizations evaluating multiple LLM providers use Experiential to A/B test models on real workloads and build data-driven routing decisions.

Considerations

Experiential is free and open-source (Apache 2.0) for the local gateway. The hosted platform at platform.experientiallabs.ai provides additional features including a dashboard for usage analytics, team management, and managed infrastructure. You supply your own provider API keys (BYOK model). The trace capture feature is experimental and currently macOS-only. Python 3.13 or newer is required for capture; the gateway itself runs on Python 3.10+. Anonymous aggregate telemetry is enabled by default but can be disabled with exp config telemetry disable.

Who It’s For

Experiential is built for teams running multiple AI coding agents who want centralized cost control, provider management, and traffic-driven optimization. It appeals to engineering leads who need visibility into agent spend across their organization and ML engineers who want to train custom models from production agent traces.