---

name: preset-skill-openai description: "Use when configuring or revising the built-in openai (OpenAI-compatible) TUI preset, including pointing it at another vendor's OpenAI-compatible API or an account pool such as sub2api / subs-pool." version: 1.0.0 last_changed_at: "2026-09-29T00:00:00Z" related_files:

• tui/internal/preset/preset.go
• tui/internal/tui/preset_editor.go
• tui/internal/tui/doctor.go
• tui/internal/tui/SKILL.md
• tui/CONTRACT.md
• tui/internal/preset/revision.go
• tui/internal/headless/preset_revision.go maintenance: "If you find stale or incorrect information here, use the lingtai-issue-report skill to assemble evidence and obtain per-issue human consent before filing an issue. Never include secrets, credentials, tokens, or private paths."

---

openai preset (OpenAI-compatible)

Use this child for the built-in openai template (display "OpenAI-compatible") and for any saved preset whose provider is openai. openaiPreset in tui/internal/preset/preset.go ships provider openai, an empty model, api_key_env OPENAI_API_KEY, base_url null, and wire_api chat_completions, with web_search on DuckDuckGo, vision inheriting the agent's own provider, and the skills default.

Template-specific settings

• base_url — optional. Empty means the official OpenAI API (https://api.openai.com/v1). Set it to any OpenAI-compatible endpoint: another vendor's compatible API, a local server, a relay, or an account pool. The value is SDK-style and carries its own version path (for example http://127.0.0.1:8080/v1); requests go to {base_url}/chat/completions or {base_url}/responses.
• wire_api — chat_completions (the default, written explicitly) or responses. With responses, responses_transport may select http (the default, omitted) or websocket.
• api_key_env — the env-var name holding the key (default OPENAI_API_KEY). Editing and saving the template stamps a fresh per-preset slot (OPENAI_1_API_KEY, OPENAI_2_API_KEY, ...) so two presets never overwrite each other's key.
• model — free text. There is no universal model: read the configured endpoint's own catalog (an authenticated GET {base_url}/models when it implements it, otherwise its documentation) and record the exact served ID. Save refuses an empty model.
• thinking — standard reasoning effort: none, minimal, low, medium, high, xhigh, or omitted (the kernel default). It is sent as reasoning effort on either wire.
• service_tier — normal (omitted) or fast (sent as priority). Whether an endpoint honors it is the endpoint's business.

Other vendors are reached here, not through vendor-named providers: pick openai, set base_url to the vendor's OpenAI-compatible endpoint, paste that vendor's key, and type the vendor's model ID. A vendor that exposes an Anthropic-compatible endpoint instead uses reference/anthropic/SKILL.md.

Subscriptions and account pools (sub2api, subs-pool)

A subscription is not an API key. To use subscriptions (for example several ChatGPT/Codex accounts) behind one endpoint, run an external pool/proxy such as sub2api or subs-pool and point an openai preset at it. The TUI and kernel do not install, start, or supervise the pool, and there is no built-in pool provider (codex-pool is retired). A pool that serves Codex models speaks the Responses wire:

"llm": {
  "provider": "openai",
  "wire_api": "responses",
  "base_url": "http://127.0.0.1:<port>/v1",
  "api_key_env": "SUBS_POOL_API_KEY",
  "model": "gpt-5.6-sol"
}

• base_url is the pool's serve address plus /v1.
• api_key_env names the env var holding the pool's local access key (the key the proxy was started with), never a ChatGPT token.
• model is a model the pooled accounts serve.

In the preset editor: open the openai template, set wire_api to responses, fill in base_url and model, and paste the pool's local access key on the API-key row. Install, account management, and quota live in the pool's own README; if the proxy is down, requests fail like any unreachable endpoint. For a single ChatGPT account, use the codex preset (reference/codex/SKILL.md) instead.

TUI surfaces to revise

Start at openaiPreset in tui/internal/preset/preset.go. In tui/internal/tui/preset_editor.go the provider row cycles the four families (editorProviders); openai has no providerModels entry, so the model row is free text; base_url is free text with the official endpoint shown when empty; wireAPIOptions, responsesTransportOptions, levelThinkingOptions, and serviceTierOptions drive the other rows, and normalizeLLMForCommit writes wire_api explicitly and drops stale transport/thinking/service-tier values and the retired api_compat field. /doctor probes this family with GET {base_url}/models (Bearer) and, on the Chat Completions wire, one max_tokens=1 completion (probeLLM in tui/internal/tui/doctor.go).

Reviewed deterministic revision

Prepare an evidence-bound manifest and explicit input, then run lingtai-tui presets revise --manifest PATH --input PATH --mode dry-run|check|apply [--output-dir PATH]. Review the JSON plan, use dry-run/check before apply, and apply only to a new explicit output directory. revision.go validates hashes, route bindings, and evidence and preserves unowned bytes; wire_api and responses_transport changes are eligible only for openai Responses targets.

Maintenance: If the relevant TUI preset/page is revised, check whether this sub-skill also needs revision and, if so, include it in the same PR.

Operations

For save, endpoint/capability, availability, activation/refresh, or troubleshooting, use the five shared operation children under reference/operations/; this child owns the OpenAI-compatible family facts, so inspect the actual saved manifest for user-owned endpoint, model, credential, and capability facts.