Open-source AI model gateway and routing control plane spanning hosted, BYOK, local and custom models, with traffic-driven optimization features.
Experiential Labs is an open-source AI model gateway and control plane that puts hosted providers, BYOK connections, local models and custom/fine-tuned models behind one endpoint. The hosted service exposes an OpenAI-compatible `/v1` API and Anthropic Messages support, while the open-source gateway can be self-hosted.12
The gateway covers practical platform controls: per-key/person/agent budgets, model allowlists, usage attribution, provider failover and routing policies. Experiential also layers optimization features over production traffic, including caching opportunities, model suggestions and optional per-prompt routing; it can also use traces to train a specialized model that the customer owns.13
Those optimization claims are differentiated, but they need careful evidence handling. Experiential publishes case studies and guarantees around cost/quality improvements; AiToolMap treats these as first-party performance evidence rather than independent validation.1
Pricing is unusually explicit. Free costs $0/month and includes 500 credits. Pro costs $20/month with 2,000 credits at 1 cent each. Enterprise uses custom committed credits. Routed tokens are billed at provider list price with 0% gateway markup, and BYOK is supported across plans.4
The open-source gateway can also run on the customer's own infrastructure and expose the core `/v1` endpoint, key management, budgets, provider waterfalls and usage API.42 That gives buyers a credible path from a free local/self-hosted deployment to a managed gateway without locking the core routing layer behind a proprietary-only service.
Experiential publishes meaningful security controls. The current security page documents ZDR-only routing, no-training-provider routing, provider allowlists and per-request attribution. It also states that the company is SOC 1 audited, with the report available on request.5
The open-source repository improves inspectability. It documents anonymous aggregate PostHog telemetry enabled by default but says telemetry excludes prompts, traces, actions, observations, paths, model names, credentials and raw customer content; telemetry can be disabled locally.2
Y Combinator independently confirms the company, founders, Summer 2026 batch and its gateway/custom-model positioning.6 That is useful identity and company-context evidence, not product-quality validation.
The fixed 50-source panel found no current exact-product professional review, user-rating signal or independent benchmark eligible for numeric transfer. Four panel domains were inaccessible through the lawful crawler path and were not bypassed. Missing evidence is treated as absent, not negative.
Experiential Labs is best suited to engineering and AI platform teams that use several model providers and need centralized policy, budgets, attribution and failover while retaining the option to self-host. It is particularly relevant when inference traffic is large enough that per-request routing, caching and model specialization could materially affect cost.3
It is less compelling for a small application with one model/provider and no governance or routing complexity. Teams considering its custom-model/optimization layer should validate the claimed quality/cost gains on their own traces before committing production traffic.
the open-source core, hosted/self-hosted flexibility, provider-price routing, transparent Free/Pro pricing, strong BYOK/local-model support, budgets/allowlists, and clear security-policy controls.45 The one-endpoint approach also makes it straightforward to repoint existing OpenAI-compatible agents and applications.2
primarily maturity and evidence depth. Experiential Labs is a 2026 company, the independent review ecosystem is effectively absent, and its most ambitious model-training and cost/quality claims come from internal tests/case studies.61 AiToolMap therefore gives the product a strong desk score but LOW confidence until independent operational evidence accumulates.
Products matched by shared tasks, audience, workflow, product type and capabilities.