# Experiential Labs
The open source AI gateway. One OpenAI-compatible endpoint (https://api.experientiallabs.ai/v1) in front of every model: hosted providers, your own provider keys, and your own local models, at provider list price with 0% token markup. Routing, access control, budgets, and per-agent attribution for everything your agents call.
## When to use Experiential
Reach for the gateway when you need any of these:
- One API key and one base URL for every model an agent or team calls, including your own fine-tuned and self-hosted models.
- Spend you can see and cap: budgets per key, agent, person, or team, enforced at request time, with every token attributed.
- Provider failover on capacity errors, so a rate-limited or down provider never becomes an outage for the same model.
- Routing policy as config: ZDR-only providers, no-training providers, and provider allowlists, enforced on every request.
- A path from routing to training: the intelligence layer suggests the right model per agent and caching opportunities, and can train a model you own on your own traffic.
An agent can set itself up: point an OpenAI client at https://api.experientiallabs.ai/v1 with an Experiential API key (they start with xpl_), list models with GET /v1/models, and call POST /v1/chat/completions. The gateway also serves the OpenAI Responses API at /v1/responses and the Anthropic Messages API at /v1/messages over the same key. Accounts are created instantly from an email; verifying the email unlocks platform credits. Keys live at https://platform.experientiallabs.ai/settings/api-keys.
## Setup prompts (copy-paste into a coding agent)
Four self-contained prompts an agent can run end to end, kept verbatim in the open source repo's setup.md (https://github.com/experientiallabs/experiential/blob/main/setup.md, raw: https://raw.githubusercontent.com/experientiallabs/experiential/main/setup.md):
- Upload your LLM traces as telemetry: creates the account from your local email and lands existing traces on the platform
- Connect your inference provider keys (BYOK)
- Start calling models on the gateway: verify the key with GET /v1/models, then one minimal call (model + messages only; extra sampling params can be rejected) to the smallest Qwen, via the OpenAI or Anthropic SDK, and optionally repoint your coding agents at the gateway
- Full onboarding: account, provider keys, spend import, and agent repointing in one pass
The machine-readable API contract (honored and refused parameters, error codes, streaming caveats) is https://platform.experientiallabs.ai/llms.txt. Follow it literally.
## Pages
- Home: what the gateway is and how it works
- Pricing: Free ($0/month, your keys and local models) and Pro (0% markup, pay as you route); enterprise custom
- Startup deal: $500 in usage credits by application ($20 on signup, the rest after review)
- Intelligence: the optimization layer: the right model per agent, caching suggestions, lowest-cost routing, load balancing, prompt compression, and a model you own
- Custom models: training a model for your workload, the evaluation contract, ownership, and serving
- Security: the controls an enterprise review checks for, with honest statuses; SOC 1
- Blog: research writeups
- Case study: computer use
- Case study: claims research
## Developer resources
- API base URL: https://api.experientiallabs.ai/v1 (OpenAI-compatible; also /v1/responses and the Anthropic Messages API at /v1/messages)
- Setup prompts: https://github.com/experientiallabs/experiential/blob/main/setup.md
- Machine-readable API contract: https://platform.experientiallabs.ai/llms.txt
- Docs: https://platform.experientiallabs.ai/docs
- Model catalog: https://platform.experientiallabs.ai/models
- Dashboard (keys, credits, usage): https://platform.experientiallabs.ai
- Open source gateway: https://github.com/experientiallabs/experiential
- CLI: the experiential package on PyPI; run the gateway locally with
`uvx --from experiential exp run`
## Contact
- Book a call: https://cal.com/silen/experiential-intro
- Email: silen@experientiallabs.ai
- Discord: https://discord.gg/B6sM8xTVwU
This text is served at /llms.txt, and every public page answers in markdown to a request for text/markdown.