Experiential Labs
Experiential Labs builds an open-source AI gateway that puts hosted providers, customer keys, local models, routing, access controls, budgets and attribution behind one OpenAI-compatible endpoint. Its intelligence layer analyzes production traces to recommend cheaper or better routes, identify caching opportunities, and train a specialized model the customer owns.
At a Glance
- AI platform and infrastructure teams
- Companies with significant GPT, Claude or other frontier-model inference bills
- Enterprise software and agent developers
- Teams in computer use, claims research, fact checking and other repetitive agent workflows
- +1 more
AI Tools by Experiential Labs
(1)
Experiential
Open Source AI Model Gateway
Discussions
No discussions yet
Be the first to start a discussion about Experiential Labs
Latest News
How we make the API fast: about 1 ms gateway overhead on the fastest lane, with a Rust data plane, fewer database round trips and an async ledger.
Failing over LLM providers without changing the model: provider waterfalls preserve model identity across outages, throttling and transport failures.
Hardened to carry all your traffic: reliability work cut request-path database round trips from seven to two and added bounded overload handling.
Claims research case study: trained Qwen3.5-4B achieved 10.9% relative higher verdict accuracy, 9.4x faster processing and 90% lower cost than Haiku 4.5 on held-out AVeriTeC claims.
Products & Services
Rust-based, self-hostable gateway and router exposing an OpenAI-compatible /v1 endpoint for hosted, BYOK, local and custom models. It provides provider waterfalls, failover, API-key management, budgets, model allowlists, attribution, request logs and a usage API.
Managed endpoint at api.experientiallabs.ai/v1 with platform-funded credits, hosted model access, provider routing and a dashboard for catalog, usage, spend, limits and request logs.
Trace-driven optimization that recommends the best model per prompt or agent, finds caching and prompt-compression opportunities, identifies batch workloads and reasoning-effort savings, and selects the lowest-cost healthy deployment.
Managed training and evaluation of smaller models on customer traces using distillation, GRPO/reinforcement learning and supervised fine-tuning; models are tested on simulations and customer evals, deployed behind the same endpoint with frontier fallback, and can be downloaded and self-hosted.
Market Position
Experiential Labs positions itself as an open-source, zero-token-markup alternative to multi-model gateways and inference routers such as LiteLLM, OpenRouter, Portkey and Helicone. Its differentiators are provider-cost pass-through, self-hosting, policy/budget/attribution controls, and a trace-to-simulation layer that can optimize routing and train a customer-owned model rather than only proxying requests.
Leadership
Founders
Kion Fallah
Co-founder and CEO; previously Staff Research Scientist at Waabi, where he led the mixed-reality simulation team. Holds a PhD in Machine Learning from Georgia Tech focused on synthetic data.
Silen Naihin
Co-founder and CTO; built and grew AutoGPT to 160,000 GitHub stars, worked on AI for science at the U.S. Department of Energy, and previously co-founded Stackwise (YC W24). His research and engineering background includes continual learning, interpretability and evaluations.
Executive Team
Kion Fallah
Co-founder and CEO
Former Staff Research Scientist at Waabi leading mixed-reality simulation; Georgia Tech ML PhD focused on synthetic data.
Silen Naihin
Co-founder and CTO
Built and grew AutoGPT to 160K GitHub stars; former AI-for-science researcher at DOE; co-founder of Stackwise; works across continual learning, interpretability and evaluations.
Founding Story
The founders saw companies repeatedly pay frontier-model prices for repetitive work while the prompts, traces and use cases generated by that spending remained an unowned asset. They started Experiential Labs to provide an open-source gateway with one key and no token markup, then use customer traffic to simulate workloads, improve routing and train smaller specialized models that customers own.
Business Model
Revenue Model
Hosted gateway usage is sold as credits, with routed tokens charged at provider list price and 0% token markup. The company also sells a $20/month Pro credit plan, top-ups at 1 cent per credit, enterprise committed-credit plans, and managed intelligence/custom-model engagements.
Pricing Tiers
500 credits per month; hosted providers, one endpoint, budgets, allowlists, attribution, top-ups and community support; 0% token markup.
2,000 credits per month at 1 cent each, with selectable allotments up to 500,000; adds BYOK and local models, intelligence features, prompt-capture opt-out, caching, model suggestions, auto routing, organization policies and dedicated support.
Committed credits at the lowest per-credit rate; SAML/SCIM, advanced RBAC, private networking, data residency, security reviews, SOC 1 report access, named contact and a customer-owned model engagement.
Target Markets
- AI platform and infrastructure teams
- Companies with significant GPT, Claude or other frontier-model inference bills
- Enterprise software and agent developers
- Teams in computer use, claims research, fact checking and other repetitive agent workflows
- Organizations needing self-hosting, BYOK, local GPUs, data-residency or zero-retention controls
- Organizations standardizing access to multiple frontier-model providers
- AI platform teams managing model access, spend, budgets, keys and attribution across users and agents
- Developers routing coding agents and production applications through one API
- Teams using BYOK, self-hosted GPUs, local models or custom deployments
- High-volume repetitive agent workloads suitable for caching, batching or smaller specialized models
- Computer-use agents operating desktop workflows
- Computer-use agent workload
- Claims-research workload on the AVeriTeC benchmark