@jacobbd/relay-ai 0.2.3 → 0.2.5

package/AGENTS.md

@@ -150,9 +150,9 @@ In all cases `process.env['OPENCODE_API_KEY']` is set immediately so the key is
 - `--restore` globs `models-*.json` (CLI) and `app-models-*.json` (App); the new files are `models-favorites.json` and `app-models-favorites.json`.
 - Zen/Go favorites are skipped in Codex (use Claude or Desktop gateway).
 
-## Release status (v0.2.3)
+## Release status (v0.2.5)
 
-Current version is **v0.2.3** — native provider registry, agent boot flags (`--provider` / `--model`), `relay-ai --ai`, Codex favorites catalog, and OpenCode cloud provider setup in `relay-ai providers add`.
+Current version is **v0.2.5** — native provider registry, agent boot flags (`--provider` / `--model`), `relay-ai --ai`, Codex favorites catalog, Codex App restore handling, reasoning capability metadata, and OpenCode cloud provider setup in `relay-ai providers add`.
 
 **Known limitations (by design):**
 - Cost display in Codex is always inaccurate for non-Anthropic models.

package/CHANGELOG.md

@@ -1,88 +1,13 @@
 # Changelog
 
-## [Unreleased]
-
-## [0.2.3] - 2026-06-15
-
-### Added
-- **Cross-surface favorites catalog** — Codex CLI (`relay-ai codex`) and Codex App (`relay-ai codex-app`) use `relay-ai models` favorites for mid-session model switching.
-- Server wizard defaults to favorites-only catalog when favorites are configured.
-- Provider labels in gateway model picker when using favorites-only mode.
-- **OpenCode Zen / Go in `relay-ai providers add`** — dedicated cloud API key path; template search for "opencode" offers the correct add flow.
-
-## [0.2.0] - 2026-06-10
+## [0.2.5] - 2026-06-18 (Official Launch Release)
 
 ### Added
-- **Native provider registry** — `relay-ai providers` command (add, import, list, remove, refresh-models)
-- Provider templates for 17+ SDK-backed providers and custom OpenAI/Anthropic endpoints
-- First-run inline wizard when registry is empty
-- Import key validation (rejects OpenCode placeholder keys; skips Vertex/Bedrock/Azure key storage)
-- Async pricing enrich with bundled `src/data/pricing-cache.json`
-- Secure trace logs under `~/.relay-ai/logs/`
-- [docs/TROUBLESHOOTING.md](docs/TROUBLESHOOTING.md)
-
-### Changed
-- **`relay-ai claude` / `models` / `server`** load from registry first — no `opencode serve` on launch
-- Removed `subscriptionTier` and config model-list cache (Zen filter lives on registry provider)
-- `--setup` now points users to `relay-ai providers`
-
-## [0.1.0] - 2026-06-09
-
-### Changed
-- **Rebrand:** `opencode-starter` → **`relay-ai`**. CLI command, config dir (`~/.relay-ai/`), and env override (`RELAY_AI_HOME`) updated throughout.
-- Migrates existing config from `~/.opencode-starter/` and legacy credential-store entries on first run.
-
-## [0.3.0] - 2026-06-07
-
-### Added
-- Local OpenCode provider discovery — launch Claude Code with any provider configured in OpenCode: Groq, Mistral, xAI, Google/Gemini, OpenAI, Anthropic-direct, Ollama, OpenRouter, Cerebras, Perplexity, Bedrock, Vertex, and more. Includes full Gemini thinking + tool calls, OpenAI Responses-API models (GPT-5.4+, GPT-5.5, Codex, o-series), and Mistral.
-- Bundled OpenCode SDK provider packages (Cerebras, Perplexity, Bedrock, Vertex, Azure, Together AI, DeepInfra, Alibaba, GitLab, Venice, and others) so any provider configured in OpenCode resolves at runtime without extra installs.
-- `relay-ai models` — interactive favorites manager (up to 20 models) for mid-session switching. With favorites set, Claude Code's `/model` switches live between your starting model and your favorites.
-- Model picker search and paginated browse for large catalogs; recent models per provider shown at the top of pickers.
-- Accurate `context_window` in synthetic `/v1/models` responses so Claude Code's status bar shows real remaining context.
-- `relay-ai server` exposes local-provider models with per-model routing; `GET /models` never returns API keys.
-
-### Changed
-- All providers route through a single Vercel AI SDK adapter (`ai` + `@ai-sdk/*`), which owns wire format, endpoint selection, and provider quirks (Gemini `thought_signature`, xAI multi-agent `/responses`, Mistral message ordering). Both the `claude` launch proxy and the `server` command use it; Anthropic-format models remain direct passthrough.
-- Local provider discovery trusts OpenCode's `api.npm` — removed the static SDK allowlist; any configured provider appears in the picker and routes through the SDK adapter (except `@ai-sdk/anthropic` passthrough and `@ai-sdk/openai-compatible` without a base URL).
-- `createLanguageModel` is async and dynamically imports the `create*` factory from whatever npm package OpenCode assigns.
-
-### Fixed
-- Local OpenAI providers: use OpenCode's `api.id` as the upstream model ID so catalog aliases (e.g. `gpt-5.5-fast` → `gpt-5.5`) don't hit `model_not_found` on the OpenAI API.
-- GPT-5.5 multi-turn reasoning: round-trip encrypted reasoning content via the Responses API without leaking SDK warnings into the Claude Code TUI.
-
-### Docs
-- Note that Claude Code persists launched models to `~/.claude/settings.json` and may cache gateway catalogs — bare `claude` can show relay-ai aliases after a session.
-- Updated README and CLAUDE.md for SDK adapter proxy naming, OpenCode-trusted discovery, `upstreamModelId`, and provider compatibility.
-
-## [0.2.5] - 2026-06-05
-
-- Fix: Pass `X-API-Key` headers to upstream servers for paid Anthropic-format models.
-- Fix: Blacklist deprecated MiMo models `mimo-v2-pro` and `mimo-v2-omni` to prevent runtime failures.
-
-## [0.2.4] - 2026-06-05
-
-- Fix: Server network mode now displays the actual local LAN IP address instead of the `<this-computer-ip>` placeholder.
-- Fix: Server password prompt now shows input in plaintext so users can verify what they typed before confirming.
-- Fix: Server network mode startup output now shows the actual server password so users can share/copy it.
-
-## [0.2.3] - 2026-06-05
-
-- Fix: The `relay-ai server` command now automatically resolves/loads the API key from the OS credential store (Keyring/Keychain/Credential Manager) if it's not exported in the shell environment.
-- Docs: Added instructions to the `README.md` on how to upgrade the package.
-
-## [0.2.2] - 2026-06-05
-
-- Simplified dry-run key-save logging in `resolveOrCollectApiKey` — replaced a 7-branch `if-else` chain with a `Record<SaveChoice, string>` lookup table. No behavioral change.
-
-## [0.2.1]
+- **Native provider registry** — Add, list, remove, refresh, and import providers with secure OS credential storage and templates for OpenRouter, Groq, Mistral, Together AI, Zen/Go, and SDK-backed custom endpoints.
+- **Claude Code and Codex launchers** — Launch registry models through `relay-ai`, including provider/model boot flags, local OpenCode provider discovery, recent models, search, pagination, and favorites catalogs for mid-session switching.
+- **Unified SDK gateway** — Route non-Anthropic providers through the Vercel AI SDK adapter while preserving Anthropic-compatible tool use, streaming, context windows, and model catalogs.
+- **Codex App integration** — Launch the desktop app with registry providers and restore Codex App model settings after interrupted sessions.
+- **Reasoning capability metadata** — Resolve reasoning controls from provider metadata, including OpenRouter `supported_parameters`, so models such as GLM 5.2 receive compatible reasoning options instead of stale defaults.
+- **Foreground server gateway** — Run `relay-ai server` for Claude Desktop or LAN usage, with registry-backed routing, password protection, and optional Vertex AI support.
+- **First-run setup** — Configure providers from an inline wizard or import existing OpenCode provider settings.
 
-- Changed the Claude Code launch path to the `relay-ai claude` command namespace. Bare `relay-ai` now prints help and migration guidance instead of launching Claude Code.
-- Preserved passthrough Claude Code args after `claude`, including `-c`, `--resume <session-id>`, session IDs, and args after `--`.
-- Added foreground `relay-ai server` mode for local or LAN API gateway use.
-- Added Anthropic-compatible and limited OpenAI-compatible server endpoints.
-- Moved app preferences and model cache to `~/.relay-ai/config.json`.
-- Added one-time migration from the previous OS-native config path.
-- Added opt-in saved server password support for network server mode.
-- Updated documentation with the supported tools command table, Claude examples, dry-run/setup/trace placement, and migration note.
-- Added MIT license file.

package/README.md

@@ -11,7 +11,7 @@
 [![License](https://img.shields.io/npm/l/@jacobbd/relay-ai)](https://github.com/jacob-bd/relay-ai/blob/main/LICENSE)
 [![Buy Me a Coffee](https://img.shields.io/badge/Buy%20Me%20a%20Coffee-FFDD00?style=flat-square&logo=buy-me-a-coffee&logoColor=black)](https://buymeacoffee.com/jacobbd)
 
-**relay-ai** is an interactive CLI that launches AI coding tools and runs Anthropic-compatible API gateways on your machine. Today that means **Claude Code** and **Claude Desktop (Cowork + Code)**. Tomorrow we'll add more agents.
+**relay-ai** is an interactive CLI that launches AI coding tools and runs local API gateways on your machine. Currently, it supports **Claude Code**, **Claude Desktop (Cowork + Code)**, the **OpenAI Codex CLI**, and the **Codex desktop app (macOS + Windows)**.
 
 Pick your backend:
 
@@ -71,7 +71,7 @@ Bare `relay-ai` prints help and migration guidance. Use `relay-ai claude` for th
 ## Prerequisites
 
 - Node.js 18+
-- A supported AI coding tool installed (e.g. [Claude Code](https://www.npmjs.com/package/@anthropic-ai/claude-code))
+- A supported AI coding tool installed (e.g. [Claude Code](https://www.npmjs.com/package/@anthropic-ai/claude-code) or [OpenAI Codex](https://www.npmjs.com/package/@openai/codex))
 - At least one provider configured via `relay-ai providers add` or `import` — **or** an [OpenCode API key](https://opencode.ai/auth) for Zen/Go cloud backends
 - [OpenCode CLI](https://opencode.ai) only if you want **one-time import** from an existing OpenCode setup (optional)
 - For **Vertex gateway:** [Google Cloud SDK](https://cloud.google.com/sdk) with `gcloud auth application-default login`, a GCP project with Vertex AI enabled, and Claude partner models enabled in that project
@@ -255,7 +255,7 @@ relay-ai server --vertex
 
 **Shorthand aliases** (for Claude Code `/model` and `settings.json`): `sonnet`, `opus`, `haiku`. Append `[1m]` for 1M context on Sonnet and Opus only (Haiku stays 200k).
 
-**Custom catalog:** copy `vertex-models.example.json` to `~/.relay-ai/vertex-models.json` and edit. Override the config directory with `RELAY_AI_HOME`.
+**Custom catalog:** copy `assets/vertex-models.example.json` to `~/.relay-ai/vertex-models.json` and edit. Override the config directory with `RELAY_AI_HOME`.
 
 When the gateway is running: