Gram describes itself as the open source stack behind Speakeasy's AI control plane, built for the platform and security engineers who have to make AI usable across a company. Its Observe surface treats token and cost telemetry as first-class, on an OpenTelemetry foundation, so a team can find an expensive session and put a budget behind it.
That matters because these are the people who must say yes or no to an agent with no cost boundary. The gateway itself is already swappable here: OpenRouter is in the tech stack, and the agent-providers list includes LiteLLM as an "open-source LLM gateway and proxy." For the same self-hosting and platform teams, that makes an additional OpenAI-compatible gateway a small, familiar choice rather than a new concept.
I'd like to propose OrcaRouter as one more optional provider. It would not replace or change any existing provider — it would just be another entry alongside what Gram already supports.
I'm an engineer on the OrcaRouter team. OrcaRouter exposes an OpenAI-compatible API and uses standard API-key authentication, so it should fit the same shape as the existing gateway boundary: server/internal/thirdparty/openrouter/unified_client.go already builds an OpenAI-compatible chat request behind a ChatClient, and the chat service talks to it through a CompletionClient interface. Upstream of that boundary, OrcaRouter could be selected much like LiteLLM already is. A few capabilities that map onto what Gram users care about:
- One OpenAI-compatible endpoint spanning many chat, reasoning, and image models, so there are fewer per-vendor base URLs and credentials to configure.
- Adaptive routing with automatic provider failover — an upstream 5xx or rate limit gets retried against healthy capacity instead of surfacing mid-session to an agent.
- Scoped API keys with budgets, usage tracking, and team access controls, which matches Gram's own per-team budget and observability model rather than duplicating it.
OrcaRouter is already present in the open-source ecosystem — for example promptfoo, Dify, NocoBase, and CAMEL — and https://www.orcarouter.ai/built-with tracks that work.
One disclosure: OrcaRouter runs an optional open-source partner program where approved projects can receive a 5% revenue share on usage attributed to their integration. Taking part is not a prerequisite for integrating, and I'm happy to follow whatever disclosure or governance rules this project has.
Would a provider like this be welcome in Gram? If it is, I'd be glad to open an implementation PR and work through the contribution guidelines with you.
Gram describes itself as the open source stack behind Speakeasy's AI control plane, built for the platform and security engineers who have to make AI usable across a company. Its Observe surface treats token and cost telemetry as first-class, on an OpenTelemetry foundation, so a team can find an expensive session and put a budget behind it.
That matters because these are the people who must say yes or no to an agent with no cost boundary. The gateway itself is already swappable here: OpenRouter is in the tech stack, and the agent-providers list includes LiteLLM as an "open-source LLM gateway and proxy." For the same self-hosting and platform teams, that makes an additional OpenAI-compatible gateway a small, familiar choice rather than a new concept.
I'd like to propose OrcaRouter as one more optional provider. It would not replace or change any existing provider — it would just be another entry alongside what Gram already supports.
I'm an engineer on the OrcaRouter team. OrcaRouter exposes an OpenAI-compatible API and uses standard API-key authentication, so it should fit the same shape as the existing gateway boundary:
server/internal/thirdparty/openrouter/unified_client.goalready builds an OpenAI-compatible chat request behind aChatClient, and the chat service talks to it through aCompletionClientinterface. Upstream of that boundary, OrcaRouter could be selected much like LiteLLM already is. A few capabilities that map onto what Gram users care about:OrcaRouter is already present in the open-source ecosystem — for example promptfoo, Dify, NocoBase, and CAMEL — and https://www.orcarouter.ai/built-with tracks that work.
One disclosure: OrcaRouter runs an optional open-source partner program where approved projects can receive a 5% revenue share on usage attributed to their integration. Taking part is not a prerequisite for integrating, and I'm happy to follow whatever disclosure or governance rules this project has.
Would a provider like this be welcome in Gram? If it is, I'd be glad to open an implementation PR and work through the contribution guidelines with you.