generated: '2026-08-27'
method: searched
source: >-
  GitHub code search across the openai organization (`org:openai extension:proto`)
  on 2026-08-27, which returned five files across three non-forked
  OpenAI-owned repositories. Each was fetched verbatim from
  raw.githubusercontent.com (HTTP 200) and saved unmodified. Fork status of every
  repository was checked via the GitHub API — none is a fork.
description: >-
  OpenAI publishes real Protobuf service definitions, and every one of them is a
  TOOLING protocol rather than an API contract. Nothing here describes
  api.openai.com: there is no gRPC surface for chat, responses, embeddings or
  administration, and none is documented. What exists is three protocols inside
  the Codex CLI (code-mode host, exec-server relay, thread config) and two inside
  OpenAI's Apple-Silicon VM orchestration stack (Orchard, tart-guest-agent).
  Recorded because they are genuine first-party published contracts under an
  OpenAI-owned org, and flagged plainly so nobody mistakes them for a gRPC
  version of the OpenAI API.
covers_platform_api: false
services:
  - file: openai-codex-code-mode-v1.proto
    package: codex.code_mode.v1
    service: CodeModeHost
    rpcs: 9
    repo: https://github.com/openai/codex
    path: codex-rs/code-mode-protocol/src/grpc/codex.code_mode.v1.proto
    description: >-
      Hosts stateful JavaScript execution for Codex and delegates nested tool
      calls to the session owner, keeping large tool inputs and outputs off the
      session event stream so independent HTTP/2 streams progress concurrently.
      Streaming on both OpenSession and SubscribeToToolCalls.
  - file: openai-codex-exec-server-relay-v1.proto
    package: codex.exec_server.relay.v1
    service: null
    rpcs: 0
    repo: https://github.com/openai/codex
    path: codex-rs/exec-server/src/proto/codex.exec_server.relay.v1.proto
    description: Message definitions only — the relay wire format for the Codex exec server. No service block.
  - file: openai-codex-thread-config-v1.proto
    package: codex.thread_config.v1
    service: ThreadConfigLoader
    rpcs: 1
    repo: https://github.com/openai/codex
    path: codex-rs/config/src/thread_config/proto/codex.thread_config.v1.proto
    description: Loads per-thread configuration for a Codex session.
  - file: openai-orchard-rpc.proto
    package: null
    service: Controller
    rpcs: 3
    repo: https://github.com/openai/orchard
    path: rpc/orchard.proto
    description: >-
      Controller RPC for Orchard, OpenAI's orchestrator for running Tart virtual
      machines across a cluster of Apple Silicon devices. Infrastructure, not product.
  - file: openai-tart-guest-agent-rpc.proto
    package: null
    service: Agent
    rpcs: 3
    repo: https://github.com/openai/tart-guest-agent
    path: proto/rpc/agent.proto
    description: Guest agent RPC for Tart VMs. Infrastructure, not product.
totals:
  files: 5
  services: 4
  rpcs: 16
buf_build:
  published: false
  note: No openai module was found on buf.build; the .proto files live only in the source repositories.