Integrations

PLLM puts a client-controlled gateway between an application and the preparation and inference services. The gateway binds to loopback, holds plaintext application state inside the trusted client boundary, and exposes:

• POST /v1/responses
• POST /v1/responses/compact
• POST /v1/chat/completions
• GET /v1/models

Preparation and inference roles do not receive plaintext prompts, tool schemas, tool arguments, or tool results. They still observe contract-specific metadata such as public model identity, tensor shapes, timing, traffic volume, and approximate sequence length. Review the trust boundary before moving either role.

Choose a path

• Local gateway starts all roles for development or connects the gateway to separately operated services.
• Responses API describes the Responses API surface and conformance evidence.
• Chat Completions API covers clients that use /v1/chat/completions.
• OpenAI Python SDK uses the official openai client.
• OpenAI Agents SDK uses its Responses API model adapter.
• Codex configures a custom Responses API provider.
• OpenCode configures an OpenAI-compatible Chat Completions API provider.

All-local mode co-locates preparation and inference. It tests transport and development workflows; it does not establish operator separation or non-collusion.