Files

hzhang 33de696653 feat: Plexum-openai-provider v0.1 — OpenAI codex CLI as backend

Plexum ProviderPlugin that wraps the local `codex` CLI binary
(OpenAI Codex CLI ≥ 0.135). Same CLI-driven pattern as
Plexum-gemini-provider and Plexum-anthropic-provider's contractor.

internal/runner/ (~200 LOC):
- Per Plexum turn, fork:
    codex exec [resume <thread_id>] \
      --skip-git-repo-check \
      --dangerously-bypass-approvals-and-sandbox \
      --json \
      [--model <m>] \
      "<last user msg>"
  in agent workspace, with stdin = /dev/null (codex would otherwise
  block waiting for additional input).
- Stream-parses JSONL events:
    thread.started        → save thread_id to .plexum-codex-session
    item.completed        → if type==agent_message, emit text_delta
    turn.completed        → capture usage (input/output/cached/reasoning)
    error / turn.failed   → emit canonical EventError
- Other event types (turn.started, item.started, reasoning, etc.)
  logged at debug, don't produce user-facing events for v1

cmd/plexum-openai-provider-plugin/ implements ProviderPluginWithAgent
(needs AgentContext.Workspace as cwd).

Model surface:
  default: ["codex"]  (no --model flag → CLI default; only thing
                       ChatGPT-account installs support)
  override via config.supported_models + config.model_args for
  API-key installs that want gpt-5 / o3 / etc.

HostConfig (all optional):
  binary           default "codex" (operator should set abs path for
                   systemd PATH)
  extra_args       appended before the prompt
  supported_models override the advertised model list
  model_args       map plexum id → CLI --model value

No api_key field — codex CLI handles auth via ~/.codex/ state
(ChatGPT login OR OPENAI_API_KEY env).

End-to-end verified against local install (ChatGPT login):

1. CLI turn 1:        "OpenAI built me."
2. CLI turn 2 (resume): "I said: 'OpenAI built me.'" (multi-turn ✓)
3. Fabric channel e2e: alice → cx agent → codex CLI → outbound REST →
   seq=23: "Use list comprehensions for concise, readable filtering and
            transformation of iterables."

Known: when daemon runs under systemd PATH doesn't include nvm/local
bin dirs; operator must set absolute binary path in config.json.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

2026-05-31 20:44:42 +01:00

2.3 KiB

Raw Permalink Blame History

Plexum-openai-provider

Plexum ProviderPlugin that wraps the local codex CLI binary (OpenAI Codex CLI ≥ 0.135). Same CLI-driven pattern as Plexum-gemini-provider and Plexum-anthropic-provider's contractor mode.

Status

v0.1: subprocess-per-turn, JSONL stream parsing, per-workspace session continuity via codex exec resume <thread_id>. Single model codex advertised by default; operator can override.

Auth

No API key in this plugin's config. The codex CLI handles auth via ~/.codex/ (ChatGPT login OR OPENAI_API_KEY). Run codex login once (or set env) and this plugin just shells out.

Install

cd ~/Plexum-openai-provider
./scripts/install.sh

Operator notes in ~/.plexum/plugins/plexum-openai-provider/config.json:

{
  "binary": "/home/<you>/.local/bin/codex",
  "extra_args": []
}

(absolute path is required when the daemon runs under systemd — systemd doesn't inherit shell PATH).

Then .plugins.allow += ["plexum-openai-provider"] and:

plexum agent-add --model codex my-agent
systemctl --user restart plexum
plexum say --agent-id my-agent --session-id $(plexum new-session --agent-id my-agent) "hello"

How it works

Per Plexum turn:

codex exec [resume <thread_id>] \
  --skip-git-repo-check \
  --dangerously-bypass-approvals-and-sandbox \
  --json \
  [--model <m>] \
  "<last user message>"

cwd = <agent workspace>, stdin = /dev/null (codex would otherwise hang waiting for additional input). Codex emits JSONL:

{"type":"thread.started","thread_id":"..."}
{"type":"turn.started"}
{"type":"item.completed","item":{"id":"...","type":"agent_message","text":"..."}}
{"type":"turn.completed","usage":{"input_tokens":...,...}}

thread.started.thread_id is captured into <workspace>/.plexum-codex-session so the next turn passes exec resume <id> instead of exec. item.completed of type agent_message becomes text_delta; turn.completed.usage populates the canonical Usage on message_end.

ChatGPT account vs API key

ChatGPT-account installs only support the default model; explicit --model gpt-5-codex etc. returns 400 invalid_request_error. With an API key install, operator can declare more model ids via config.supported_models + config.model_args.

License

Same as Plexum.

2.3 KiB Raw Permalink Blame History