experiential: An open source gateway and router for agent workflows.
What problem it solves
Agent workflows need to use hosted, BYOK, and local models while applying access, usage, and spending controls. They also need a way to turn production traffic into routing decisions or a specialized model tailored to observed workloads.
Key features
Unified model access — Route hosted, BYOK, and local models through one OpenAI-compatible API, with support for connecting provider credentials.
Source: README
Access and spending controls — Define which users and agents can access particular models and use cases, and apply command budgets to limit spending.
Source: README
Traffic-based optimization — Collect agent traces, simulate routing decisions, and use observed traffic to build a project-specific router or fine-tune an open source model.
Source: README
Local and hosted operation — Run a local gateway for agent workflows or use the hosted platform, which exposes compatible OpenAI and Anthropic API surfaces.
Source: README
Quick start
Copied verbatim from the project's README.
pip install experiential
expimport exp
with exp.load_router("my-project") as client:
response = client.chat.completions.create(
model="my-project",
messages=[{"role": "user", "content": "hello"}],
)Who it may suit
Inference: This may suit teams and developers operating agent workflows that need unified model access, routing controls, traffic collection, or project-specific model optimization.
Topics to discuss
Which access and spending policies should be available when controlling users, agents, models, and use cases?
What OpenTelemetry trace schema should teams standardize when preparing traffic for router optimization?
Which hosted, BYOK, and local model providers should be prioritized for the gateway and onboarding experience?
Written by ReadmeX AI from the project's README, source and public GitHub data; it may contain mistakes.
Read the original README https://github.com/experientiallabs/experiential#readme
ReadmeX opened this community for a trending project; it is not run by the project. Maintainers can claim it from the community page.
Comments
…