OGX
Open GenAI Stack — a drop-in OpenAI-compatible agentic server. Formerly Llama Stack.
Open-source agentic API server, renamed from Llama Stack to OGX along with a shift to being model-agnostic rather than Llama-first. A drop-in replacement for the OpenAI API that runs on a laptop, a datacenter or the cloud: chat completions, embeddings, the Responses API with server-side tool calling and MCP, vector stores, files, batches and versioned skill bundles. Because it speaks the OpenAI wire format, Trustvice attaches as an LLM proxy rather than an SDK.
Beta
Status
LLM proxy
Pattern
Python
Runtime
Not yet published
SDK version
Readiness
What “Beta” means for OGX
The integration path is built and tested. Real customers can attach now, but expect minor breaking changes until the 1.0 contract freeze.
How it attaches
LLM proxy.
Point OGX's model base URL at the Trustvice proxy. Trustvice records the full request + response and forwards to the real provider; streaming is preserved.
Replace your agent's LLM base URL with the Trustvice proxy URL. Trustvice captures the request and forwards to the real provider.
- 1
Pick an auth mode
Workspace-key mode: send Authorization: Bearer twk_… and let Trustvice attach your stored provider credentials. Passthrough mode: send your real provider key and add X-Trustvice-Workspace for routing.
- 2
Change the base URL
Set `the provider's `url` in the run.yaml provider config` to the matching Trustvice proxy URL (…/api/proxy/openai, …/api/proxy/anthropic, or …/api/proxy/gemini). OGX speaks the OpenAI-compatible wire format, so use the matching proxy path.
- 3
Verify
The first request through the proxy shows up immediately in Witness. The synchronous path adds ≤50 ms; capture happens asynchronously after the response is returned.
{
"model": {
"provider": "openai-compatible",
"url": "https://api.trustvice.com/proxy/openai",
"model": "gpt-4o",
"apiKey": "twk_abc123"
}
}Coverage
2 event types captured.
agent.invocationEvery top-level agent run — the input, the resolved output, timing, and token usage.
tool.callEach tool / function the agent invokes, with arguments and the returned result.
The full pattern guide
LLM proxy — end to end.
Every framework on this pattern shares the same platform-side spine, endpoints, and gotchas. The pattern guide covers them in full.
Read the LLM proxy guideNeed help landing OGX?
Talk to the integration team.
We help you land the integration end to end — including the parts that live in your environment.
Request integration help
Trustvice 360声明规则，让每次运行都接受检查。
智能体、证据、治理、商务和披露都在 Trustvice 360 中。免费注册，在一个仪表板中打开所有套件。