Show HN: We built open OpenRouter that distills usage into a better model

Experiential Labs has unveiled an open-source platform called OpenRouter, designed to streamline agent workflows and distill usage into a more efficient model. This platform functions as an open-source gateway and router for agent workflows, enabling users to start a local OpenAI-compatible gateway on their first run.

During the initial setup, the wizard prompts users to select a provider connection, model, and reasoning effort, persisting each chosen connection before displaying defaults for the public alias, identity, and a $50.00 command budget. A unique one-time key is then generated, which users can capture and use to send a request.

For those who prefer a managed gateway, Experiential Labs offers a hosted platform at platform.experientiallabs.ai, which mirrors the OpenAI-compatible (and Anthropic Messages) API at https://api.experientiallabs.ai/v1.

Using a Python environment, users can load a fitted project router as an official OpenAI client backed by its own private gateway. This can be achieved by starting the local gateway with 'exp' or 'exp run', and then loading the compiled native data plane that serves every route on loopback.

To enhance the functionality of OpenRouter, users are encouraged to collect OpenTelemetry traces from their current agent. If users are short on time, they can utilize the public terminal-tasks OTLP dataset. Once traces are collected, users can build a project, during which the build command will guide them through providers, models, and budget, and request a trace file.

PostHog product telemetry is enabled by default in OpenRouter, but it is important to note that anonymous aggregate telemetry is employed, ensuring that prompts, traces, actions, observations, paths, model names, credentials, and raw customer content are not included. This preference is stored locally in .exp/settings.toml.

For more information on the repository and documentation, users can refer to AGENTS.md. OpenRouter is an open-source model gateway that provides a unified control plane across closed, open-source, local, and custom models.