Experiential Labs
Open source AI gateway turning traffic into a better model
Open-source OpenAI-compatible gateway over hosted providers, BYOK, and local models at provider list price with 0% token markup. Routing, access control, budgets and per-agent attribution sit in the request path. Production traces can train a custom router or a specialized model you own. Hosted API at platform.experientiallabs.ai.
What is Experiential Labs?
Experiential is an open-source AI gateway. You point an OpenAI-compatible (or Anthropic Messages) client at one endpoint and reach hosted providers, your own keys, and local models at provider list price with no token markup. It adds routing, access control, budgets and per-agent attribution, and can turn production traces into a custom router or a specialized model you own.
Key features
- One OpenAI-compatible API (`/v1/chat/completions`, `/v1/responses`) plus Anthropic Messages at `/v1/messages` over the same key
- BYOK pass-through for OpenAI, Anthropic, Gemini, Azure, Bedrock, Fireworks and OpenRouter at provider list price with 0% token markup
- Self-hosted local gateway (`pip install experiential` / `uvx --from experiential exp run`) and a hosted platform at platform.experientiallabs.ai
- Access control: which users and agents can use which models, for which use cases, and how much they can spend
- Per-agent attribution of calls
- Copy-paste setup prompts for Claude Code, Cursor, Codex and similar, so a coding agent can finish the wiring
- Trace ingest (OpenTelemetry JSONL) to build a project router or fine-tune a model you own
- Apache-2.0 license; keys on the hosted platform start with `xpl_`
Who it's for
- A team running Claude Code, Cursor or Codex against several providers who wants one base URL, one budget, and no extra tax per token
- A company that will upload existing LLM traces and train a cheaper router for the repetitive slice of traffic
- A developer who wants to self-host the gateway on loopback and keep provider keys on their machine
- An agent that can list `/v1/models` and call chat completions with a single Experiential key
When not to use it
Not a drop-in if you only needed a multi-provider proxy — LiteLLM already does that. The training-from-traffic loop needs traces and a budget you are willing to spend on fitting. The marketing root experientiallabs.ai currently shows a research page, so onboarding is GitHub + platform.experientiallabs.ai, not a single polished marketing site.
FAQ
How much does Experiential cost?
The gateway is Apache-2.0 and you can self-host it for free. Hosted BYOK is pass-through at provider list price with 0% token markup, per the GitHub README. Paid hosted tiers are not itemized first-party. The YC profile claims a 50% cost SLA on full deployments — that is a vendor claim, not a public price list.
Which APIs does the gateway speak?
OpenAI Chat Completions and Responses, and Anthropic Messages, on the same key. Locally you start the gateway with `exp` / `exp run`. Hosted base URL is https://api.experientiallabs.ai/v1.
Will it train on my prompts by default?
You opt in by uploading or exporting traces. Anonymous PostHog product telemetry is on by default in the CLI and, per the README, never includes prompts, traces, actions, observations, paths, model names, credentials, or raw customer content. You can disable it in `.exp/settings.toml`.
Can a coding agent set it up?
Yes. SETUP.md is a set of copy-paste prompts for Claude Code, Cursor, Codex and similar. The agent creates an account from email, connects keys, and repoints the OpenAI client at the gateway.
Share this launch
Embed this badge
<a href="https://orangebot.ai/product/experiential-labs" target="_blank" rel="noopener noreferrer"> <img src="https://orangebot.ai/api/badge/experiential-labs.svg" alt="Featured on OrangeBot" width="200" height="54" /> </a>