mu-agent 0.2.3 → 0.2.5

package/CHANGELOG.md

@@ -2,6 +2,30 @@
 
 ## [Unreleased]
 
+## [0.2.5] - 2026-10-24
+
+**0.2.5 removes Laya from mu's built-in judges. A configuration that still names it switches to mude 1.0 or Jev on its own.**
+
+### Changed
+
+- A configuration that still names Laya keeps working. `laya` in `tiers` or a route, and a judge of type `local`, are read without it: mude 1.0 takes its place when it is installed on this Mac and its key (`MU_JUDGE_MUDE_KEY`) is set, otherwise the judges left do, or Jev when none is left. A route left with no judge goes back to `tiers`. The first session with a terminal or the desktop app (not `mu --print` or json mode) writes that to mu.json once. It changes only those values, keeps a copy in `~/.mu/backups`, and says so once: `judge.notice` with the code `laya_removed` and the `judge`, then the line itself. A mu.json that isn't plain JSON is left alone. `MU_JUDGE=laya` is skipped in every session that has it. For such a configuration the launcher starts mude's server in the background, as it does for `mude`.
+- `/mu judge laya` and `/mu route <point> laya` say that Laya is no longer built in. An `http` judge needs a `baseUrl`; its client is `HttpJudgeProvider` (it was `LocalJudgeProvider`).
+
+### Removed
+
+- Laya, the third-party local judge, is no longer built in: the `laya` judge, the `local` judge type, its sidecar (`kyrn/bin/kyrn-judge-local`, `kyrn/local-judge/server.py` and its lock, no longer in the package) and `mu judge setup | start | stop | status | run`. Those commands now say so and point to `mu judge mude install` and `mu judge model add <link>`; `mu judge` alone prints the judge commands. The launcher no longer starts a sidecar, and `mu doctor` drops its rows.
+- `continuation.localPromised`, which only Laya used, and the one-question-at-a-time path of a `local` judge: `tool.admission` and `tool.injection` always ask in batches, and the questions about a message start together.
+
+## [0.2.4] - 2026-10-24
+
+**0.2.4 brings Gemini 4 Argon through Google Antigravity and fixes third-party OpenAI-compatible providers that turned away the `developer` role.**
+
+### Fixed
+
+- Signed in with Google Antigravity, mu now lists Gemini 4 Argon (`gemini-4-argon-low`, `-medium` and `-high`) and the Gemini 3.8 Flash tiers, and can use them. mu introduced itself to Antigravity as its IDE, version 1.107.0 (`antigravity/1.107.0 darwin/arm64`). Google lists and serves those models only to the Antigravity command line from 1.3.0, and answered 404 for them. mu now introduces itself as the command line (`antigravity/cli/1.4.0`; `MU_ANTIGRAVITY_VERSION` still raises the version). Gemini 4 Argon thinks at low, medium or high, never off. A placeholder for a model the account may not use, such as `gemini-4-argon-disabled`, is left out. The context Antigravity gives a model (1 million tokens for Claude Opus and Sonnet 5.5) is now used.
+- An account that was already signed in to Antigravity gets the new models without signing in again. Its list was read only at sign-in; now the terminal and the desktop app also read it in the background as a session starts. Print and JSON runs don't, nor does `PI_OFFLINE`. Signing out still drops the account's models, these included.
+- A custom OpenAI-compatible provider (`"api": "openai-completions"` in `models.json`) no longer fails with "unknown role developer" for a thinking model. Only `api.openai.com`, Azure OpenAI and Xiaomi MiMo get `developer` messages and `max_completion_tokens`; other endpoints get `system`, `max_tokens` and no `store`. When an endpoint still rejects the `developer` role, the max tokens field, `reasoning_effort`, `stream_options`, `store` or the prompt cache fields with a 400 (or a gateway's 5xx wrapping one), mu sends the request once more without it, before any output, and remembers that for the model until it restarts. `compat` in `models.json` overrides the defaults.
+
 ## [0.2.3] - 2026-10-24
 
 **0.2.3 adds `mu judge model` for decision models from Hugging Face, makes mu decide when a mid-run message reaches the work, and makes Sign in with ChatGPT cope with a busy port 1455.**

package/README.md

@@ -27,13 +27,13 @@ mu is built on [pi](https://github.com/earendil-works/pi). It hands the small qu
 
 ## Choose a judge
 
-Out of the box, the judge is Jev 1.13, TypeSafe's hosted judge model, free on OpenCode Zen for a limited time and with no key needed. What it judges goes to OpenCode, which doesn't train on it, and mu reminds you of that once a day. To send nothing anywhere, run `mu setup --judge off` (or choose "Not now" when `mu setup` asks): no decision point asks a judge, and mu's features keep working on their rules. Or use the local judge, Laya.
+Out of the box, the judge is Jev 1.13, TypeSafe's hosted judge model, free on OpenCode Zen for a limited time and with no key needed. What it judges goes to OpenCode, which doesn't train on it, and mu reminds you of that once a day. To send nothing anywhere, run `mu setup --judge off` (or choose "Not now" when `mu setup` asks): no decision point asks a judge, and mu's features keep working on their rules. Or, on a Mac with Apple silicon, use mude 1.0 with `--local-only`.
 
 - **Jev** (hosted): set `TYPESAFE_API_KEY` (TypeSafe), `MU_JUDGE_OPENROUTER_API_KEY` (Jev on OpenRouter), `AI_GATEWAY_API_KEY` (Vercel AI Gateway), `OPENCODE_API_KEY` (OpenCode Zen) or `CLOUDFLARE_API_KEY` with `CLOUDFLARE_ACCOUNT_ID` (Cloudflare Workers AI) in your environment, or add it to `~/.mu/.env` as `KEY=value`. `mu setup` asks for a key and writes it there. mu reads that file as data, never prints it and starts using the key automatically.
 - **Another classifier** from pi's catalog: `/mu judge classifier:<provider>/<model>`, for example Cloudflare's Clef (`clef`, `clef-flash`) or a System One model on OpenRouter, with that provider's key.
 - **CLM** (self-hosted; its encoder needs a GPU): start `clm-serve` from [Contrastive-LM/CLM](https://github.com/Contrastive-LM/CLM), then run `/mu judge clm`. For a server on another machine, add `"judges": { "clm": { "type": "clm", "baseUrl": "http://<host>:8700" } }` to `~/.mu/agent/mu.json`. If the server was started with `CLM_API_KEY`, set the same value in `MU_JUDGE_CLM_API_KEY`. `mu doctor` checks the server.
 - **mude 1.0** (local, a Mac with Apple silicon, macOS 14+), mu's own decision model, optimized for coding: `mu judge mude install` downloads it (1.8 GB), sets it up in `~/.mu/local-judge/mude`, starts its server on 127.0.0.1 and puts it first. What it can't answer goes to the judges you had before, which are hosted; `mu judge mude enable --local-only` keeps everything on your Mac. Its server uses about 4 GB of memory until `mu judge mude stop`.
-- **Laya** (local, macOS on Apple silicon): `mu judge setup` installs it into `~/.mu/local-judge`. It downloads about 930 MB, and only when you run that command.
+- **Other decision models from Hugging Face** (local, a Mac with Apple silicon): `mu judge model add <link>` adds one that is packaged like mude 1.0 and makes it the judge. Laya is no longer included: a `mu.json` that still names it gets mude 1.0 in its place when mude is set up, otherwise the remaining judges or Jev, and mu says so once.
 - **Any model you already use:** `/mu judge llm:<provider>/<model>`.
 
 34 of the 40 decision points act on their verdicts from the start (`active`). To only record what the judge would do, run `/mu mode default shadow`, or put `"modes": { "default": "shadow" }` in `~/.mu/agent/mu.json`.
@@ -47,7 +47,7 @@ mu -p "prompt"           run one prompt, print the answer and exit
 mu -c | -r               continue the last session | pick a session to resume
 mu setup                 connect a model and choose a judge (see mu setup --help for scripting and services)
 mu judge mude <cmd>      mude 1.0 on this Mac: install | enable | disable | start | stop | status | uninstall
-mu judge <cmd>           manage the local judge, Laya (macOS on Apple silicon): setup | start | stop | status | run
+mu judge model <cmd>     other decision models from Hugging Face: add <link> | list | use | remove
 mu ledger [n] [--json]   print the judge's verdicts from the last n sessions
 mu import <file>...      import Claude Code, Codex and ZCode conversations; --list shows them (mu import --help)
 mu doctor                check the installation, the judges and the connections