coderouter-cli 1.8.3__tar.gz → 1.8.5__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/CHANGELOG.md +64 -3
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/PKG-INFO +11 -7
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/README.en.md +12 -7
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/README.md +10 -6
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/doctor.py +63 -8
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/free-tier-guide.en.md +4 -1
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/free-tier-guide.md +3 -1
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/hf-ollama-models.md +17 -3
- coderouter_cli-1.8.5/docs/llamacpp-direct.en.md +232 -0
- coderouter_cli-1.8.5/docs/llamacpp-direct.md +285 -0
- coderouter_cli-1.8.5/docs/lmstudio-direct.en.md +405 -0
- coderouter_cli-1.8.5/docs/lmstudio-direct.md +407 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/quickstart.en.md +15 -7
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/quickstart.md +12 -6
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/examples/providers.yaml +181 -2
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/pyproject.toml +1 -1
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_setup_sh.py +6 -6
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/.gitignore +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/LICENSE +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/__init__.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/__main__.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/adapters/__init__.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/adapters/anthropic_native.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/adapters/base.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/adapters/openai_compat.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/adapters/registry.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/cli.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/cli_stats.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/config/__init__.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/config/capability_registry.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/config/env_file.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/config/loader.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/config/schemas.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/data/__init__.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/data/model-capabilities.yaml +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/doctor_apply.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/env_security.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/errors.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/ingress/__init__.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/ingress/anthropic_routes.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/ingress/app.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/ingress/dashboard_routes.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/ingress/metrics_routes.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/ingress/openai_routes.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/logging.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/metrics/__init__.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/metrics/collector.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/metrics/prometheus.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/output_filters.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/routing/__init__.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/routing/auto_router.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/routing/capability.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/routing/fallback.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/translation/__init__.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/translation/anthropic.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/translation/convert.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/coderouter/translation/tool_repair.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/assets/dashboard-demo.png +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/designs/v1.5-dashboard-mockup.html +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/designs/v1.6-auto-router-verification.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/designs/v1.6-auto-router.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/openrouter-roster/README.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/openrouter-roster/latest.json +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/retrospectives/v0.4.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/retrospectives/v0.5-verify.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/retrospectives/v0.5.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/retrospectives/v0.6.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/retrospectives/v0.7.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/retrospectives/v1.0-verify.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/retrospectives/v1.0.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/security.en.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/security.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/troubleshooting.en.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/troubleshooting.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/usage-guide.en.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/usage-guide.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/when-do-i-need-coderouter.en.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/docs/when-do-i-need-coderouter.md +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/examples/.env.example +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/examples/providers.auto-custom.yaml +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/examples/providers.auto.yaml +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/examples/providers.note-2026.yaml +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/examples/providers.nvidia-nim.yaml +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/scripts/demo_traffic.sh +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/scripts/openrouter_roster_diff.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/scripts/verify_v0_5.sh +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/scripts/verify_v1_0.sh +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/__init__.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/conftest.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_adapter_anthropic.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_auto_router.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_capability.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_capability_degraded_payload.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_capability_registry.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_claude_code_suitability.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_cli.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_cli_stats.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_config.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_dashboard_endpoint.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_doctor.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_doctor_apply.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_env_file.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_env_security.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_errors.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_examples_yaml.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_fallback.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_fallback_anthropic.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_fallback_cache_control.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_fallback_misconfig_warn.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_fallback_paid_gate.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_fallback_thinking.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_ingress_anthropic.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_ingress_profile.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_metrics_collector.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_metrics_endpoint.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_metrics_jsonl.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_metrics_prometheus.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_openai_compat.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_openrouter_roster_diff.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_output_filters.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_output_filters_adapters.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_reasoning_strip.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_tool_repair.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_translation_anthropic.py +0 -0
- {coderouter_cli-1.8.3 → coderouter_cli-1.8.5}/tests/test_translation_reverse.py +0 -0
|
@@ -6,6 +6,55 @@ versioning follows [SemVer](https://semver.org/).
|
|
|
6
6
|
|
|
7
7
|
---
|
|
8
8
|
|
|
9
|
+
## [v1.8.5] — 2026-04-28 (doctor NEEDS_TUNING メッセージを v1.8.3 thinking-aware budget の事実に揃える + `docs/lmstudio-direct.md` 新規)
|
|
10
|
+
|
|
11
|
+
**Theme: 文言の整合 patch + ドキュメント補完。**v1.8.3 で `tool_calls` / `num_ctx` / `streaming` の 3 probe に thinking-aware budget (256 / 1024) を入れた。今回はその事実を NEEDS_TUNING 時の detail メッセージに反映し、operator が「probe budget が小さすぎたのでは」と疑う余地をなくす。あわせて v1.8.4 で実機検証した LM Studio 0.4.12 経由経路を `docs/llamacpp-direct.md` と対をなす形で `docs/lmstudio-direct.md` (+ `.en.md`) として正式化。
|
|
12
|
+
|
|
13
|
+
- Tests: 737 → 737 (既存 assert は phrase-substring を見ていないので追従不要、新規 assertion は不足分を 1 件追加)
|
|
14
|
+
- Runtime deps: 5 → 5 (22 sub-release 連続据え置き)
|
|
15
|
+
- Backward compat: 完全互換、`providers.yaml` / `~/.coderouter/model-capabilities.yaml` / コード側 API 変更なし
|
|
16
|
+
|
|
17
|
+
### Changes
|
|
18
|
+
|
|
19
|
+
#### Doctor NEEDS_TUNING 文言更新 (suggestion を thinking-aware budget 前提に揃える)
|
|
20
|
+
|
|
21
|
+
- **`coderouter/doctor.py` `_probe_tool_calls`**: 「Common for quantized small models」を残しつつ、thinking モデル時は `Probed with thinking-aware budget (1024 tokens, covers reasoning_content plus the call) — this is a true tools=false case, not budget exhaustion.` を前置。非 thinking 時は `Probed with default budget (256 tokens) — the model produced no tool-shaped output at all.` を前置
|
|
22
|
+
- **`coderouter/doctor.py` `_probe_streaming`**: `finish_reason='length'` 偽陽性回避のため、thinking 時は `Probe sent max_tokens=1024 (thinking-aware), so the cap is server-side options.num_predict rather than the probe budget.` を前置。非 thinking 時は `Probe sent max_tokens=512;` 系を前置
|
|
23
|
+
- **`coderouter/doctor.py` `_probe_num_ctx`**: 「canary missing」3 ケース (declared=None / declared<threshold / declared>=threshold) すべてに、thinking モデル時は `Probe sent max_tokens=1024 (thinking-aware), so the miss is prompt-side truncation rather than reply truncation.` の budget note を追加。これで operator が「probe の reply budget が足りなかったのでは」という疑問を即座に消せる
|
|
24
|
+
|
|
25
|
+
#### Documentation 補完: `docs/lmstudio-direct.md` 新規
|
|
26
|
+
|
|
27
|
+
- **`docs/lmstudio-direct.md` / `.en.md` 新規** — v1.8.4 で実機検証した LM Studio 0.4.12 経由経路を `docs/llamacpp-direct.md` と対をなす形で 7 step + Troubleshooting で。M3 Max 64GB / Q4_K_M / Metal 想定 + GUI 操作前提の canonical recipe
|
|
28
|
+
- Step 1: LM Studio install & Discover タブで Q4_K_M モデルダウンロード (Qwen3.5 9B / Qwen3.6 35B-A3B / Jackrong/Qwopus3.5-9B-v3-GGUF)
|
|
29
|
+
- Step 2: Chat タブで Load Model (Context 32768 / GPU max / Flash Attention ON)
|
|
30
|
+
- Step 3: Local Server タブで Port 1234 / Just-in-time Model Loading: ON / Start Server
|
|
31
|
+
- Step 4: curl 直叩き (OpenAI 互換 + Anthropic 互換 両ルート、native tool_calls / native tool_use 両方確認)
|
|
32
|
+
- Step 5: CodeRouter に provider 登録 (`kind: openai_compat` 経路 + `kind: anthropic` 経路の 2 種)
|
|
33
|
+
- Step 6: doctor 6 probe で動作確認 (両ルートとも全 probe OK)
|
|
34
|
+
- Step 7: CodeRouter 経由 end-to-end (Anthropic prompt caching `cache_read_input_tokens: 280` 観測も含む)
|
|
35
|
+
|
|
36
|
+
### Why
|
|
37
|
+
|
|
38
|
+
v1.8.3 で `tool_calls` probe の active-harmful 誤診断 (thinking モデルに対して `tools: false` 提案) を fix したが、メッセージ文面はそのまま v1.8.2 以前の言い回し (「Common for quantized small models」のみ) を残していた。operator が NEEDS_TUNING を見たときに「probe budget が小さすぎたのでは」「v1.8.2 のバグの再発では」と疑う余地が文面上残っていたのを、**実装が既に thinking-aware なので断定できる** という事実に文言を揃える。診断ツールの出力は実装の confidence を反映すべき。
|
|
39
|
+
|
|
40
|
+
`docs/lmstudio-direct.md` は v1.8.4 で実機検証 + `examples/providers.yaml` に provider 例追加までは済ませていたが、`docs/llamacpp-direct.md` と並ぶレベルの canonical recipe ドキュメントが欠けていた。LM Studio 経由が現時点で最も `qwen35` / `qwen35moe` architecture を安定して動かせる経路 (Anthropic prompt caching まで透過) なので、operator が辿り着けるドキュメントとして正式化。
|
|
41
|
+
|
|
42
|
+
### Migration
|
|
43
|
+
|
|
44
|
+
`pyproject.toml version 1.8.3 → 1.8.5`、`coderouter --version` は 1.8.5 を返す。**手元の `~/.coderouter/providers.yaml` は触らない限り完全に変化なし**。doctor 出力の文面が変わるが verdict と suggested_patch の semantic は完全互換。
|
|
45
|
+
|
|
46
|
+
### Files touched
|
|
47
|
+
|
|
48
|
+
```
|
|
49
|
+
M CHANGELOG.md
|
|
50
|
+
M coderouter/doctor.py
|
|
51
|
+
M pyproject.toml
|
|
52
|
+
A docs/lmstudio-direct.md
|
|
53
|
+
A docs/lmstudio-direct.en.md
|
|
54
|
+
```
|
|
55
|
+
|
|
56
|
+
---
|
|
57
|
+
|
|
9
58
|
## [v1.8.3] — 2026-04-26 (tool_calls probe も thinking モデル対応 + adapter で `reasoning_content` strip — llama.cpp 直叩き対応)
|
|
10
59
|
|
|
11
60
|
**Theme: v1.8.2 と同日リリースの第 2 弾 patch。Qwen3.6:35b-a3b on llama.cpp の実機検証で発見した 2 つの追加課題 — `tool_calls` probe の thinking モデル偽陽性 + llama.cpp が emit する `reasoning_content` フィールドの adapter strip 不足 — を解消。**
|
|
@@ -70,6 +119,18 @@ M tests/test_doctor.py
|
|
|
70
119
|
M tests/test_reasoning_strip.py
|
|
71
120
|
```
|
|
72
121
|
|
|
122
|
+
### Post-release docs followup (同 commit ではなく追加 commit で)
|
|
123
|
+
|
|
124
|
+
llama.cpp 直叩き経路を canonical な救済路として正式採用したのを受け、関連ドキュメントを v1.8.3 後に整理:
|
|
125
|
+
|
|
126
|
+
- **`docs/llamacpp-direct.md` / `.en.md` 新規** — `llama.cpp` build → Unsloth GGUF → `llama-server` → CodeRouter 接続を 7 step + Troubleshooting で。M3 Max 64GB / Q4_K_M / Metal 想定の canonical recipe
|
|
127
|
+
- **`setup.sh`**: 48 GB+ tier の推奨を旧 `qwen3.6:35b` → `gemma4:26b` に変更 (Ollama 経由詰みのため)。upgrade hint からも Qwen3.6 系を撤去、代わりに `docs/llamacpp-direct.md` への誘導を追加
|
|
128
|
+
- **`docs/quickstart.md` / `.en.md`**: 「より良いモデル」セクションの `ollama pull qwen3.6:35b` を撤去、`docs/llamacpp-direct.md` への誘導追加
|
|
129
|
+
- **`docs/hf-ollama-models.md`**: `ollama pull qwen3.6:35b` を「⚠️ Qwen3.6 系は Ollama 経由で詰みやすい」警告に置換、llama.cpp 直叩き経路の案内を追加
|
|
130
|
+
- **`README.md` / `.en.md`**: ドキュメント目次に「llama.cpp 直叩きガイド」行を追加、英語版言語スイッチャーにも `llama.cpp direct` リンクを追加
|
|
131
|
+
- **`examples/providers.yaml`**: `llamacpp-qwen3-6-35b-a3b` provider 例を追加 + `coding` profile chain primary に組み込み (詳細コメント付き)。Qwen3.6 系 Ollama 経路のコメントも v1.8.3 結果反映で更新
|
|
132
|
+
- **`tests/test_setup_sh.py`**: 48 GB / 64 GB tier の expected_model assertion を `qwen3.6:35b` → `gemma4:26b` に追従更新
|
|
133
|
+
|
|
73
134
|
---
|
|
74
135
|
|
|
75
136
|
## [v1.8.2] — 2026-04-26 (doctor probe を thinking モデル対応に — Gemma 4 偽陽性の解消)
|
|
@@ -399,7 +460,7 @@ $ curl -sI "https://pypi.org/pypi/coderouter-cli/json" | head -1
|
|
|
399
460
|
HTTP/2 200
|
|
400
461
|
```
|
|
401
462
|
|
|
402
|
-
CDN 伝播後に `uvx coderouter-cli --version` で本物の PyPI
|
|
463
|
+
CDN 伝播後に `uvx --from coderouter-cli coderouter --version` で本物の PyPI 経由インストールも確認済み (uv 0.11+ では package 名 ≠ executable 名のとき `--from` 必須、Issue #10 で報告者から fb)。
|
|
403
464
|
|
|
404
465
|
### Migration
|
|
405
466
|
|
|
@@ -409,8 +470,8 @@ CDN 伝播後に `uvx coderouter-cli --version` で本物の PyPI 経由イン
|
|
|
409
470
|
# 旧 (引き続き有効)
|
|
410
471
|
uv tool install --from git+https://github.com/zephel01/CodeRouter.git coderouter-cli
|
|
411
472
|
|
|
412
|
-
# 新 (PyPI から、コマンド 1 行)
|
|
413
|
-
uvx coderouter-cli serve --port 8088
|
|
473
|
+
# 新 (PyPI から、コマンド 1 行 — uv 0.11+ canonical 形式)
|
|
474
|
+
uvx --from coderouter-cli coderouter serve --port 8088
|
|
414
475
|
# あるいは恒久的に:
|
|
415
476
|
uv tool install coderouter-cli
|
|
416
477
|
```
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: coderouter-cli
|
|
3
|
-
Version: 1.8.
|
|
3
|
+
Version: 1.8.5
|
|
4
4
|
Summary: Local-first, free-first, fallback-built-in LLM router. Claude Code / OpenAI compatible.
|
|
5
5
|
Project-URL: Homepage, https://github.com/zephel01/CodeRouter
|
|
6
6
|
Project-URL: Repository, https://github.com/zephel01/CodeRouter
|
|
@@ -60,7 +60,7 @@ Description-Content-Type: text/markdown
|
|
|
60
60
|
<p align="center">
|
|
61
61
|
<a href="https://github.com/zephel01/CodeRouter/actions/workflows/ci.yml"><img src="https://github.com/zephel01/CodeRouter/actions/workflows/ci.yml/badge.svg?branch=main" alt="CI"></a>
|
|
62
62
|
<a href=""><img src="https://img.shields.io/badge/status-stable-brightgreen" alt="status"></a>
|
|
63
|
-
<a href=""><img src="https://img.shields.io/badge/version-1.8.
|
|
63
|
+
<a href=""><img src="https://img.shields.io/badge/version-1.8.5-blue" alt="version"></a>
|
|
64
64
|
<a href=""><img src="https://img.shields.io/badge/python-3.12%2B-blue" alt="python"></a>
|
|
65
65
|
<a href=""><img src="https://img.shields.io/badge/runtime%20deps-5-brightgreen" alt="deps"></a>
|
|
66
66
|
<a href=""><img src="https://img.shields.io/badge/license-MIT-yellow" alt="license"></a>
|
|
@@ -99,11 +99,13 @@ Description-Content-Type: text/markdown
|
|
|
99
99
|
| **無料で回す** | [無料枠ガイド](./docs/free-tier-guide.md) | NVIDIA NIM 40 req/min × OpenRouter 無料枠の使い分け・live 検証済みモデル表・地雷 5 点 |
|
|
100
100
|
| **要るか判断する** | [要否判定ガイド](./docs/when-do-i-need-coderouter.md) | エージェント × モデルの詳細マトリクスで「そもそも自分に必要か」を決める |
|
|
101
101
|
| **詰まったとき** | [トラブルシューティング](./docs/troubleshooting.md) | `doctor` の使い方、`.env` の export 必須、Ollama サイレント失敗 5 症状、Claude Code 連携の罠 |
|
|
102
|
+
| **llama.cpp 直叩き** | [llama.cpp 直叩きガイド](./docs/llamacpp-direct.md) | Qwen3.6 を Ollama 詰みから救出する経路。`llama.cpp` build → Unsloth GGUF → `llama-server` → CodeRouter 接続を 7 step で(v1.8.3 実機検証済)|
|
|
103
|
+
| **LM Studio 直接** | [LM Studio 直接ガイド](./docs/lmstudio-direct.md) | `qwen35` / `qwen35moe` を救う第 2 経路。LM Studio 0.4.12+ Local Server 経由で OpenAI 互換 + Anthropic 互換 (`/v1/messages`) 両対応、prompt caching 透過(v1.8.4 実機検証済)|
|
|
102
104
|
| **安全に使う** | [セキュリティ方針](./docs/security.md) | 脅威モデル・秘密情報の扱い・脆弱性報告経路 |
|
|
103
|
-
| **履歴** | [CHANGELOG](./CHANGELOG.md) | 全リリース履歴(最新: v1.8.
|
|
105
|
+
| **履歴** | [CHANGELOG](./CHANGELOG.md) | 全リリース履歴(最新: v1.8.5 — doctor NEEDS_TUNING メッセージを v1.8.3 thinking-aware budget の事実に揃える + `docs/lmstudio-direct.md` 新規) |
|
|
104
106
|
| **設計を追う** | [plan.md](./plan.md) | 設計不変項・マイルストーン・今後のロードマップ |
|
|
105
107
|
|
|
106
|
-
English versions: [Quickstart](./docs/quickstart.en.md) · [Usage guide](./docs/usage-guide.en.md) · [Free-tier guide](./docs/free-tier-guide.en.md) · [When you need it](./docs/when-do-i-need-coderouter.en.md) · [Troubleshooting](./docs/troubleshooting.en.md) · [Security](./docs/security.en.md)
|
|
108
|
+
English versions: [Quickstart](./docs/quickstart.en.md) · [Usage guide](./docs/usage-guide.en.md) · [Free-tier guide](./docs/free-tier-guide.en.md) · [When you need it](./docs/when-do-i-need-coderouter.en.md) · [Troubleshooting](./docs/troubleshooting.en.md) · [llama.cpp direct](./docs/llamacpp-direct.en.md) · [LM Studio direct](./docs/lmstudio-direct.en.md) · [Security](./docs/security.en.md)
|
|
107
109
|
|
|
108
110
|
## CodeRouter で何が楽になるか
|
|
109
111
|
|
|
@@ -184,7 +186,9 @@ curl -fsSL https://raw.githubusercontent.com/zephel01/CodeRouter/main/examples/p
|
|
|
184
186
|
> ~/.coderouter/providers.yaml
|
|
185
187
|
|
|
186
188
|
# 2. uvx で起動 (インストール + 起動が 1 行)
|
|
187
|
-
|
|
189
|
+
# PyPI 配布名 (coderouter-cli) と console script 名 (coderouter) が異なるため、
|
|
190
|
+
# uv 0.11+ では --from 形式が必須 (旧 uv でも動く canonical 形式)
|
|
191
|
+
uvx --from coderouter-cli coderouter serve --port 8088
|
|
188
192
|
```
|
|
189
193
|
|
|
190
194
|
恒久的にインストールしておきたい場合:
|
|
@@ -257,7 +261,7 @@ CodeRouter 自体は純 Python 3.12+ で、実質的な OS 対応範囲は `min(
|
|
|
257
261
|
|
|
258
262
|
**リリース単位の詳細が欲しい?** v0.x と v1.0-A/B/C の各スライス — 何が入り、何本のテストが増え、なぜ必要だったのか — は [CHANGELOG.md](./CHANGELOG.md) に揃っています。設計の不変項と今後のロードマップは [plan.md](./plan.md)。
|
|
259
263
|
|
|
260
|
-
**次の予定**(v1.0 は [plan.md §10](./plan.md)、v1.0+ は §18): v1.5 ✅ メトリクス / `/dashboard` / `coderouter stats` TUI / `scripts/demo_traffic.sh`、v1.6 ✅ `auto_router` (task-aware routing) + NVIDIA NIM 無料枠 + トラブルシュートドキュメント分離 + `--env-file` / `doctor --check-env`、v1.7 ✅ PyPI 公開 (`uvx coderouter-cli`)、v1.8 ✅ 用途別 4 プロファイル (multi/coding/general/reasoning) + Gemma 4 / Qwen3.6 / Z.AI (GLM) 登録 + `setup.sh` onboarding ウィザード + `coderouter doctor --check-model --apply` (非破壊 YAML 書き戻し) + `claude_code_suitability` startup チェック + Trusted Publishing 自動化。残り (v1.9 候補) は `coderouter doctor --network` (CI 用) / launcher スクリプト / 起動時アップデートチェック (opt-in)。
|
|
264
|
+
**次の予定**(v1.0 は [plan.md §10](./plan.md)、v1.0+ は §18): v1.5 ✅ メトリクス / `/dashboard` / `coderouter stats` TUI / `scripts/demo_traffic.sh`、v1.6 ✅ `auto_router` (task-aware routing) + NVIDIA NIM 無料枠 + トラブルシュートドキュメント分離 + `--env-file` / `doctor --check-env`、v1.7 ✅ PyPI 公開 (`uvx --from coderouter-cli coderouter`)、v1.8 ✅ 用途別 4 プロファイル (multi/coding/general/reasoning) + Gemma 4 / Qwen3.6 / Z.AI (GLM) 登録 + `setup.sh` onboarding ウィザード + `coderouter doctor --check-model --apply` (非破壊 YAML 書き戻し) + `claude_code_suitability` startup チェック + Trusted Publishing 自動化。残り (v1.9 候補) は `coderouter doctor --network` (CI 用) / launcher スクリプト / 起動時アップデートチェック (opt-in)。
|
|
261
265
|
|
|
262
266
|
### Claude Code と一緒に使う
|
|
263
267
|
|
|
@@ -432,7 +436,7 @@ suggested patch for ~/.coderouter/providers.yaml:
|
|
|
432
436
|
- v1.0 ✅ — 14 ケースのリグレッションスイート、Code Mode (スリム版 Claude Code ハーネス); 出力クリーニングは **v1.0-A** で `output_filters` チェーンとして完了
|
|
433
437
|
- v1.5 ✅ — **メトリクスダッシュボード(出荷済み)** — `MetricsCollector` + `GET /metrics.json` + `GET /metrics` (Prometheus) + `GET /dashboard` (HTML 1 ページ) + `coderouter stats` curses TUI + `scripts/demo_traffic.sh` トラフィックジェネレータ + `display_timezone` 設定
|
|
434
438
|
- v1.6 ✅ — `auto_router` (task-aware routing、`default_profile: auto` で画像/コード濃度/その他を自動振り分け) + NVIDIA NIM 無料枠 8 段チェーン + ドキュメント言語スワップ (JA primary) + トラブルシュート独立ドキュメント + `--env-file` / `doctor --check-env`
|
|
435
|
-
- v1.7 ✅ — PyPI 公開 (`uvx coderouter-cli` で 1 行起動) + Trusted Publishing 経路 (release.yml で自動 publish)
|
|
439
|
+
- v1.7 ✅ — PyPI 公開 (`uvx --from coderouter-cli coderouter` で 1 行起動) + Trusted Publishing 経路 (release.yml で自動 publish)
|
|
436
440
|
- v1.8 ✅ — **用途別 4 プロファイル + GLM/Gemma 4/Qwen3.6 公式化 + apply 自動化**: `multi` (default) / `coding` / `general` / `reasoning` の 4 プロファイル + 全プロファイルに `append_system_prompt` で Claude 風応答 nudge + `mode_aliases` (default/fast/vision/think/cheap)、Ollama 公式 tag 化された `gemma4:e4b/26b/31b` / `qwen3.6:27b/35b` を active stanza に格上げ、Z.AI を OpenAI-compat で 2 base_url 提供 (Coding Plan / General API)、`coderouter doctor --check-model --apply` で YAML パッチを非破壊書き戻し (`ruamel.yaml` round-trip でコメント・key 順序保持、冪等)、`setup.sh` onboarding ウィザード、`claude_code_suitability` startup チェック (Llama-3.3-70B 系を `claude-code-*` profile で WARN)。残り (v1.9 以降): `coderouter doctor --network` (CI 用)、launcher スクリプト (`.command` / `.sh` / `.bat`)、opt-in 起動時アップデートチェック
|
|
437
441
|
|
|
438
442
|
## `kind: openai_compat` と `kind: anthropic` の選び方
|
|
@@ -20,7 +20,7 @@
|
|
|
20
20
|
<p align="center">
|
|
21
21
|
<a href="https://github.com/zephel01/CodeRouter/actions/workflows/ci.yml"><img src="https://github.com/zephel01/CodeRouter/actions/workflows/ci.yml/badge.svg?branch=main" alt="CI"></a>
|
|
22
22
|
<a href=""><img src="https://img.shields.io/badge/status-stable-brightgreen" alt="status"></a>
|
|
23
|
-
<a href=""><img src="https://img.shields.io/badge/version-1.8.
|
|
23
|
+
<a href=""><img src="https://img.shields.io/badge/version-1.8.5-blue" alt="version"></a>
|
|
24
24
|
<a href=""><img src="https://img.shields.io/badge/python-3.12%2B-blue" alt="python"></a>
|
|
25
25
|
<a href=""><img src="https://img.shields.io/badge/runtime%20deps-5-brightgreen" alt="deps"></a>
|
|
26
26
|
<a href=""><img src="https://img.shields.io/badge/license-MIT-yellow" alt="license"></a>
|
|
@@ -58,11 +58,13 @@
|
|
|
58
58
|
| **Run for free** | [Free-tier guide](./docs/free-tier-guide.en.md) | Stacking NVIDIA NIM (40 req/min) with OpenRouter free: live-verified roster and five footguns |
|
|
59
59
|
| **Decide if you need it** | [Decision guide](./docs/when-do-i-need-coderouter.en.md) | Agent × model matrix to figure out whether CodeRouter fits your setup at all |
|
|
60
60
|
| **When stuck** | [Troubleshooting](./docs/troubleshooting.en.md) | How to use `doctor`, why `.env` needs `export`, the 5 Ollama silent-fail symptoms, Claude Code integration gotchas |
|
|
61
|
+
| **llama.cpp direct** | [llama.cpp direct guide](./docs/llamacpp-direct.en.md) | Rescue path for Qwen3.6 (Ollama is brittle). 7-step recipe: `llama.cpp` build → Unsloth GGUF → `llama-server` → CodeRouter wiring. Real-machine verified in v1.8.3. |
|
|
62
|
+
| **LM Studio direct** | [LM Studio direct guide](./docs/lmstudio-direct.en.md) | Second rescue path for `qwen35` / `qwen35moe`. LM Studio 0.4.12+ Local Server with both OpenAI-compatible and Anthropic-compatible (`/v1/messages`) routes — prompt caching survives end-to-end. Real-machine verified in v1.8.4. |
|
|
61
63
|
| **Operate safely** | [Security](./docs/security.en.md) | Threat model, secret handling, vulnerability reporting |
|
|
62
|
-
| **History** | [CHANGELOG](./CHANGELOG.md) | All releases (latest: v1.8.
|
|
64
|
+
| **History** | [CHANGELOG](./CHANGELOG.md) | All releases (latest: v1.8.5 — doctor NEEDS_TUNING messages aligned with v1.8.3 thinking-aware budget + new `docs/lmstudio-direct.md`) |
|
|
63
65
|
| **Track the design** | [plan.md](./plan.md) | Design invariants, milestones, roadmap |
|
|
64
66
|
|
|
65
|
-
日本語版: [Quickstart](./docs/quickstart.md) · [利用ガイド](./docs/usage-guide.md) · [無料枠ガイド](./docs/free-tier-guide.md) · [要否判定](./docs/when-do-i-need-coderouter.md) · [トラブルシューティング](./docs/troubleshooting.md) · [Security](./docs/security.md)
|
|
67
|
+
日本語版: [Quickstart](./docs/quickstart.md) · [利用ガイド](./docs/usage-guide.md) · [無料枠ガイド](./docs/free-tier-guide.md) · [要否判定](./docs/when-do-i-need-coderouter.md) · [トラブルシューティング](./docs/troubleshooting.md) · [LM Studio 直接](./docs/lmstudio-direct.md) · [Security](./docs/security.md)
|
|
66
68
|
|
|
67
69
|
## What gets easier with CodeRouter
|
|
68
70
|
|
|
@@ -142,8 +144,11 @@ mkdir -p ~/.coderouter
|
|
|
142
144
|
curl -fsSL https://raw.githubusercontent.com/zephel01/CodeRouter/main/examples/providers.yaml \
|
|
143
145
|
> ~/.coderouter/providers.yaml
|
|
144
146
|
|
|
145
|
-
# 2. Install + run, in one line
|
|
146
|
-
|
|
147
|
+
# 2. Install + run, in one line.
|
|
148
|
+
# The PyPI distribution name (coderouter-cli) differs from the console
|
|
149
|
+
# script name (coderouter), so uv 0.11+ requires the --from form. (It
|
|
150
|
+
# also works on older uv, so this is the canonical incantation.)
|
|
151
|
+
uvx --from coderouter-cli coderouter serve --port 8088
|
|
147
152
|
```
|
|
148
153
|
|
|
149
154
|
Want a permanent install?
|
|
@@ -216,7 +221,7 @@ What CodeRouter can do for you today:
|
|
|
216
221
|
|
|
217
222
|
**Want the per-release detail?** Every v0.x and v1.0-A/B/C slice — what shipped, how many tests it added, why it was needed — is in [CHANGELOG.md](./CHANGELOG.md). Design invariants and the forward roadmap live in [plan.md](./plan.md).
|
|
218
223
|
|
|
219
|
-
**Coming next** (see [plan.md §10](./plan.md) for v1.0, §18 for v1.0+): v1.5 ✅ metrics / `/dashboard` / `coderouter stats` TUI / `scripts/demo_traffic.sh`. v1.6 ✅ `auto_router` (task-aware routing) + NVIDIA NIM free tier + troubleshooting doc split + `--env-file` / `doctor --check-env`. v1.7 ✅ PyPI publish (`uvx coderouter-cli`). v1.8 ✅ Use-case-aware 4 profiles (multi/coding/general/reasoning) + Gemma 4 / Qwen3.6 / Z.AI (GLM) registration + `setup.sh` wizard + `coderouter doctor --check-model --apply` (non-destructive YAML write-back) + `claude_code_suitability` startup check + Trusted Publishing automation. Remaining for v1.9+: `coderouter doctor --network` (CI), launcher scripts, opt-in update check.
|
|
224
|
+
**Coming next** (see [plan.md §10](./plan.md) for v1.0, §18 for v1.0+): v1.5 ✅ metrics / `/dashboard` / `coderouter stats` TUI / `scripts/demo_traffic.sh`. v1.6 ✅ `auto_router` (task-aware routing) + NVIDIA NIM free tier + troubleshooting doc split + `--env-file` / `doctor --check-env`. v1.7 ✅ PyPI publish (`uvx --from coderouter-cli coderouter`). v1.8 ✅ Use-case-aware 4 profiles (multi/coding/general/reasoning) + Gemma 4 / Qwen3.6 / Z.AI (GLM) registration + `setup.sh` wizard + `coderouter doctor --check-model --apply` (non-destructive YAML write-back) + `claude_code_suitability` startup check + Trusted Publishing automation. Remaining for v1.9+: `coderouter doctor --network` (CI), launcher scripts, opt-in update check.
|
|
220
225
|
|
|
221
226
|
### Use it with Claude Code
|
|
222
227
|
|
|
@@ -393,7 +398,7 @@ Coming next (see [plan.md §10](./plan.md) for v1.0, §18 for v1.0+):
|
|
|
393
398
|
- v1.0 ✅ — 14-case regression suite, Code Mode (slim Claude Code harness); output cleaning shipped in **v1.0-A** (`output_filters` chain, done)
|
|
394
399
|
- v1.5 ✅ — **Metrics dashboard (shipped)** — `MetricsCollector` + `GET /metrics.json` + `GET /metrics` (Prometheus) + `GET /dashboard` (HTML one-pager) + `coderouter stats` curses TUI + `scripts/demo_traffic.sh` traffic generator + `display_timezone` config
|
|
395
400
|
- v1.6 ✅ — `auto_router` (task-aware routing; `default_profile: auto` dispatches by image attachment / code-fence ratio / else) + NVIDIA NIM free-tier 8-step chain + doc language swap (JA primary) + troubleshooting page split + `--env-file` / `doctor --check-env`
|
|
396
|
-
- v1.7 ✅ — PyPI publish (`uvx coderouter-cli` one-line bootstrap) + Trusted Publishing path (release.yml auto-publish on tag push)
|
|
401
|
+
- v1.7 ✅ — PyPI publish (`uvx --from coderouter-cli coderouter` one-line bootstrap) + Trusted Publishing path (release.yml auto-publish on tag push)
|
|
397
402
|
- v1.8 ✅ — **Use-case-aware 4 profiles + GLM/Gemma 4/Qwen3.6 official tags + apply automation**: `multi` (default) / `coding` / `general` / `reasoning` profiles + `append_system_prompt` per profile to nudge non-Claude models toward Claude-style replies + `mode_aliases` (default/fast/vision/think/cheap), Ollama-official `gemma4:e4b/26b/31b` and `qwen3.6:27b/35b` promoted to active stanzas, Z.AI provided as OpenAI-compat with two base URLs (Coding Plan / General API), `coderouter doctor --check-model --apply` writes YAML patches non-destructively (`ruamel.yaml` round-trip preserves comments + key order; idempotent), `setup.sh` onboarding wizard, `claude_code_suitability` startup check (Llama-3.3-70B in `claude-code-*` profiles emits a structured WARN). Remaining for v1.9+: `coderouter doctor --network` (CI-friendly), launcher scripts (`.command` / `.sh` / `.bat`), opt-in startup update check
|
|
398
403
|
|
|
399
404
|
## Choosing `kind: openai_compat` vs `kind: anthropic`
|
|
@@ -19,7 +19,7 @@
|
|
|
19
19
|
<p align="center">
|
|
20
20
|
<a href="https://github.com/zephel01/CodeRouter/actions/workflows/ci.yml"><img src="https://github.com/zephel01/CodeRouter/actions/workflows/ci.yml/badge.svg?branch=main" alt="CI"></a>
|
|
21
21
|
<a href=""><img src="https://img.shields.io/badge/status-stable-brightgreen" alt="status"></a>
|
|
22
|
-
<a href=""><img src="https://img.shields.io/badge/version-1.8.
|
|
22
|
+
<a href=""><img src="https://img.shields.io/badge/version-1.8.5-blue" alt="version"></a>
|
|
23
23
|
<a href=""><img src="https://img.shields.io/badge/python-3.12%2B-blue" alt="python"></a>
|
|
24
24
|
<a href=""><img src="https://img.shields.io/badge/runtime%20deps-5-brightgreen" alt="deps"></a>
|
|
25
25
|
<a href=""><img src="https://img.shields.io/badge/license-MIT-yellow" alt="license"></a>
|
|
@@ -58,11 +58,13 @@
|
|
|
58
58
|
| **無料で回す** | [無料枠ガイド](./docs/free-tier-guide.md) | NVIDIA NIM 40 req/min × OpenRouter 無料枠の使い分け・live 検証済みモデル表・地雷 5 点 |
|
|
59
59
|
| **要るか判断する** | [要否判定ガイド](./docs/when-do-i-need-coderouter.md) | エージェント × モデルの詳細マトリクスで「そもそも自分に必要か」を決める |
|
|
60
60
|
| **詰まったとき** | [トラブルシューティング](./docs/troubleshooting.md) | `doctor` の使い方、`.env` の export 必須、Ollama サイレント失敗 5 症状、Claude Code 連携の罠 |
|
|
61
|
+
| **llama.cpp 直叩き** | [llama.cpp 直叩きガイド](./docs/llamacpp-direct.md) | Qwen3.6 を Ollama 詰みから救出する経路。`llama.cpp` build → Unsloth GGUF → `llama-server` → CodeRouter 接続を 7 step で(v1.8.3 実機検証済)|
|
|
62
|
+
| **LM Studio 直接** | [LM Studio 直接ガイド](./docs/lmstudio-direct.md) | `qwen35` / `qwen35moe` を救う第 2 経路。LM Studio 0.4.12+ Local Server 経由で OpenAI 互換 + Anthropic 互換 (`/v1/messages`) 両対応、prompt caching 透過(v1.8.4 実機検証済)|
|
|
61
63
|
| **安全に使う** | [セキュリティ方針](./docs/security.md) | 脅威モデル・秘密情報の扱い・脆弱性報告経路 |
|
|
62
|
-
| **履歴** | [CHANGELOG](./CHANGELOG.md) | 全リリース履歴(最新: v1.8.
|
|
64
|
+
| **履歴** | [CHANGELOG](./CHANGELOG.md) | 全リリース履歴(最新: v1.8.5 — doctor NEEDS_TUNING メッセージを v1.8.3 thinking-aware budget の事実に揃える + `docs/lmstudio-direct.md` 新規) |
|
|
63
65
|
| **設計を追う** | [plan.md](./plan.md) | 設計不変項・マイルストーン・今後のロードマップ |
|
|
64
66
|
|
|
65
|
-
English versions: [Quickstart](./docs/quickstart.en.md) · [Usage guide](./docs/usage-guide.en.md) · [Free-tier guide](./docs/free-tier-guide.en.md) · [When you need it](./docs/when-do-i-need-coderouter.en.md) · [Troubleshooting](./docs/troubleshooting.en.md) · [Security](./docs/security.en.md)
|
|
67
|
+
English versions: [Quickstart](./docs/quickstart.en.md) · [Usage guide](./docs/usage-guide.en.md) · [Free-tier guide](./docs/free-tier-guide.en.md) · [When you need it](./docs/when-do-i-need-coderouter.en.md) · [Troubleshooting](./docs/troubleshooting.en.md) · [llama.cpp direct](./docs/llamacpp-direct.en.md) · [LM Studio direct](./docs/lmstudio-direct.en.md) · [Security](./docs/security.en.md)
|
|
66
68
|
|
|
67
69
|
## CodeRouter で何が楽になるか
|
|
68
70
|
|
|
@@ -143,7 +145,9 @@ curl -fsSL https://raw.githubusercontent.com/zephel01/CodeRouter/main/examples/p
|
|
|
143
145
|
> ~/.coderouter/providers.yaml
|
|
144
146
|
|
|
145
147
|
# 2. uvx で起動 (インストール + 起動が 1 行)
|
|
146
|
-
|
|
148
|
+
# PyPI 配布名 (coderouter-cli) と console script 名 (coderouter) が異なるため、
|
|
149
|
+
# uv 0.11+ では --from 形式が必須 (旧 uv でも動く canonical 形式)
|
|
150
|
+
uvx --from coderouter-cli coderouter serve --port 8088
|
|
147
151
|
```
|
|
148
152
|
|
|
149
153
|
恒久的にインストールしておきたい場合:
|
|
@@ -216,7 +220,7 @@ CodeRouter 自体は純 Python 3.12+ で、実質的な OS 対応範囲は `min(
|
|
|
216
220
|
|
|
217
221
|
**リリース単位の詳細が欲しい?** v0.x と v1.0-A/B/C の各スライス — 何が入り、何本のテストが増え、なぜ必要だったのか — は [CHANGELOG.md](./CHANGELOG.md) に揃っています。設計の不変項と今後のロードマップは [plan.md](./plan.md)。
|
|
218
222
|
|
|
219
|
-
**次の予定**(v1.0 は [plan.md §10](./plan.md)、v1.0+ は §18): v1.5 ✅ メトリクス / `/dashboard` / `coderouter stats` TUI / `scripts/demo_traffic.sh`、v1.6 ✅ `auto_router` (task-aware routing) + NVIDIA NIM 無料枠 + トラブルシュートドキュメント分離 + `--env-file` / `doctor --check-env`、v1.7 ✅ PyPI 公開 (`uvx coderouter-cli`)、v1.8 ✅ 用途別 4 プロファイル (multi/coding/general/reasoning) + Gemma 4 / Qwen3.6 / Z.AI (GLM) 登録 + `setup.sh` onboarding ウィザード + `coderouter doctor --check-model --apply` (非破壊 YAML 書き戻し) + `claude_code_suitability` startup チェック + Trusted Publishing 自動化。残り (v1.9 候補) は `coderouter doctor --network` (CI 用) / launcher スクリプト / 起動時アップデートチェック (opt-in)。
|
|
223
|
+
**次の予定**(v1.0 は [plan.md §10](./plan.md)、v1.0+ は §18): v1.5 ✅ メトリクス / `/dashboard` / `coderouter stats` TUI / `scripts/demo_traffic.sh`、v1.6 ✅ `auto_router` (task-aware routing) + NVIDIA NIM 無料枠 + トラブルシュートドキュメント分離 + `--env-file` / `doctor --check-env`、v1.7 ✅ PyPI 公開 (`uvx --from coderouter-cli coderouter`)、v1.8 ✅ 用途別 4 プロファイル (multi/coding/general/reasoning) + Gemma 4 / Qwen3.6 / Z.AI (GLM) 登録 + `setup.sh` onboarding ウィザード + `coderouter doctor --check-model --apply` (非破壊 YAML 書き戻し) + `claude_code_suitability` startup チェック + Trusted Publishing 自動化。残り (v1.9 候補) は `coderouter doctor --network` (CI 用) / launcher スクリプト / 起動時アップデートチェック (opt-in)。
|
|
220
224
|
|
|
221
225
|
### Claude Code と一緒に使う
|
|
222
226
|
|
|
@@ -391,7 +395,7 @@ suggested patch for ~/.coderouter/providers.yaml:
|
|
|
391
395
|
- v1.0 ✅ — 14 ケースのリグレッションスイート、Code Mode (スリム版 Claude Code ハーネス); 出力クリーニングは **v1.0-A** で `output_filters` チェーンとして完了
|
|
392
396
|
- v1.5 ✅ — **メトリクスダッシュボード(出荷済み)** — `MetricsCollector` + `GET /metrics.json` + `GET /metrics` (Prometheus) + `GET /dashboard` (HTML 1 ページ) + `coderouter stats` curses TUI + `scripts/demo_traffic.sh` トラフィックジェネレータ + `display_timezone` 設定
|
|
393
397
|
- v1.6 ✅ — `auto_router` (task-aware routing、`default_profile: auto` で画像/コード濃度/その他を自動振り分け) + NVIDIA NIM 無料枠 8 段チェーン + ドキュメント言語スワップ (JA primary) + トラブルシュート独立ドキュメント + `--env-file` / `doctor --check-env`
|
|
394
|
-
- v1.7 ✅ — PyPI 公開 (`uvx coderouter-cli` で 1 行起動) + Trusted Publishing 経路 (release.yml で自動 publish)
|
|
398
|
+
- v1.7 ✅ — PyPI 公開 (`uvx --from coderouter-cli coderouter` で 1 行起動) + Trusted Publishing 経路 (release.yml で自動 publish)
|
|
395
399
|
- v1.8 ✅ — **用途別 4 プロファイル + GLM/Gemma 4/Qwen3.6 公式化 + apply 自動化**: `multi` (default) / `coding` / `general` / `reasoning` の 4 プロファイル + 全プロファイルに `append_system_prompt` で Claude 風応答 nudge + `mode_aliases` (default/fast/vision/think/cheap)、Ollama 公式 tag 化された `gemma4:e4b/26b/31b` / `qwen3.6:27b/35b` を active stanza に格上げ、Z.AI を OpenAI-compat で 2 base_url 提供 (Coding Plan / General API)、`coderouter doctor --check-model --apply` で YAML パッチを非破壊書き戻し (`ruamel.yaml` round-trip でコメント・key 順序保持、冪等)、`setup.sh` onboarding ウィザード、`claude_code_suitability` startup チェック (Llama-3.3-70B 系を `claude-code-*` profile で WARN)。残り (v1.9 以降): `coderouter doctor --network` (CI 用)、launcher スクリプト (`.command` / `.sh` / `.bat`)、opt-in 起動時アップデートチェック
|
|
396
400
|
|
|
397
401
|
## `kind: openai_compat` と `kind: anthropic` の選び方
|
|
@@ -836,6 +836,22 @@ async def _probe_num_ctx(
|
|
|
836
836
|
)
|
|
837
837
|
|
|
838
838
|
# Canary missing → truncation occurred.
|
|
839
|
+
#
|
|
840
|
+
# v1.8.5: with the v1.8.3 thinking-aware response budget already
|
|
841
|
+
# applied (max_tokens=1024 for reasoning models — see
|
|
842
|
+
# ``_NUM_CTX_PROBE_MAX_TOKENS_THINKING``), a missing canary cannot
|
|
843
|
+
# be blamed on an under-sized reply. The fault is genuinely on the
|
|
844
|
+
# prompt side: the upstream truncated the input before the model
|
|
845
|
+
# could see the canary token at the head. This sharpens the
|
|
846
|
+
# remediation — we are confident bumping ``num_ctx`` is the right
|
|
847
|
+
# fix, not bumping the response budget.
|
|
848
|
+
thinking = _is_reasoning_model(provider, resolved)
|
|
849
|
+
budget_note = (
|
|
850
|
+
f" Probe sent max_tokens={max_tokens} (thinking-aware), so the "
|
|
851
|
+
"miss is prompt-side truncation rather than reply truncation."
|
|
852
|
+
if thinking
|
|
853
|
+
else ""
|
|
854
|
+
)
|
|
839
855
|
if declared is None:
|
|
840
856
|
return ProbeResult(
|
|
841
857
|
name="num_ctx",
|
|
@@ -845,7 +861,7 @@ async def _probe_num_ctx(
|
|
|
845
861
|
"upstream truncated the prompt. No `extra_body.options.num_ctx` "
|
|
846
862
|
"is declared, so Ollama is running at its 2048-token default, "
|
|
847
863
|
"which cannot hold Claude Code's system + tool prompts "
|
|
848
|
-
"(plan.md §9.4 symptom #1)."
|
|
864
|
+
f"(plan.md §9.4 symptom #1).{budget_note}"
|
|
849
865
|
),
|
|
850
866
|
target_file="providers.yaml",
|
|
851
867
|
suggested_patch=_patch_providers_yaml_num_ctx(provider.name, 32768),
|
|
@@ -857,7 +873,8 @@ async def _probe_num_ctx(
|
|
|
857
873
|
detail=(
|
|
858
874
|
f"canary missing — declared num_ctx={declared} is below "
|
|
859
875
|
f"the {_NUM_CTX_ADEQUATE_THRESHOLD}-token threshold needed "
|
|
860
|
-
"for Claude Code prompts. Bump it (plan.md §9.4 symptom
|
|
876
|
+
f"for Claude Code prompts. Bump it (plan.md §9.4 symptom "
|
|
877
|
+
f"#1).{budget_note}"
|
|
861
878
|
),
|
|
862
879
|
target_file="providers.yaml",
|
|
863
880
|
suggested_patch=_patch_providers_yaml_num_ctx(provider.name, 32768),
|
|
@@ -875,7 +892,7 @@ async def _probe_num_ctx(
|
|
|
875
892
|
"declared value, or the upstream is silently capping it — "
|
|
876
893
|
"verify with the model card / server logs. The suggested "
|
|
877
894
|
"patch still emits 32768 as a starting point; dial down if "
|
|
878
|
-
"the host is memory-constrained."
|
|
895
|
+
f"the host is memory-constrained.{budget_note}"
|
|
879
896
|
),
|
|
880
897
|
target_file="providers.yaml",
|
|
881
898
|
suggested_patch=_patch_providers_yaml_num_ctx(provider.name, 32768),
|
|
@@ -1041,6 +1058,24 @@ async def _probe_streaming(
|
|
|
1041
1058
|
# mid-word". Since we're already Ollama-shape-gated, the
|
|
1042
1059
|
# remediation is always the ``extra_body.options.num_predict``
|
|
1043
1060
|
# bump.
|
|
1061
|
+
#
|
|
1062
|
+
# v1.8.5: with v1.8.3's thinking-aware probe budget already
|
|
1063
|
+
# applied (max_tokens=1024 for reasoning models), a length cap
|
|
1064
|
+
# here cannot be blamed on the probe budget — the upstream is
|
|
1065
|
+
# the one capping. Surface the budget used so the operator can
|
|
1066
|
+
# rule it out at a glance.
|
|
1067
|
+
thinking = _is_reasoning_model(provider, resolved)
|
|
1068
|
+
budget_note = (
|
|
1069
|
+
f"Probe sent max_tokens={max_tokens} (thinking-aware), so "
|
|
1070
|
+
"the cap is server-side `options.num_predict` rather than "
|
|
1071
|
+
"the probe budget."
|
|
1072
|
+
if thinking
|
|
1073
|
+
else (
|
|
1074
|
+
f"Probe sent max_tokens={max_tokens}; the cap is "
|
|
1075
|
+
"server-side `options.num_predict` rather than the "
|
|
1076
|
+
"probe budget."
|
|
1077
|
+
)
|
|
1078
|
+
)
|
|
1044
1079
|
return ProbeResult(
|
|
1045
1080
|
name="streaming",
|
|
1046
1081
|
verdict=ProbeVerdict.NEEDS_TUNING,
|
|
@@ -1048,9 +1083,9 @@ async def _probe_streaming(
|
|
|
1048
1083
|
f"stream closed with `finish_reason='length'` after only "
|
|
1049
1084
|
f"{len(content)} chars (expected ≥ "
|
|
1050
1085
|
f"{_STREAMING_PROBE_MIN_EXPECTED_CHARS}). Upstream is "
|
|
1051
|
-
"capping output — most likely `options.num_predict`. "
|
|
1052
|
-
"Bump it via `extra_body` (plan.md §9.4
|
|
1053
|
-
"streaming variant)."
|
|
1086
|
+
f"capping output — most likely `options.num_predict`. "
|
|
1087
|
+
f"{budget_note} Bump it via `extra_body` (plan.md §9.4 "
|
|
1088
|
+
"symptom #1 streaming variant)."
|
|
1054
1089
|
),
|
|
1055
1090
|
target_file="providers.yaml",
|
|
1056
1091
|
suggested_patch=_patch_providers_yaml_num_predict(
|
|
@@ -1239,13 +1274,33 @@ async def _probe_tool_calls(
|
|
|
1239
1274
|
|
|
1240
1275
|
# Nothing tool-shaped at all.
|
|
1241
1276
|
if declared:
|
|
1277
|
+
# v1.8.5: with the v1.8.3 thinking-aware budget already applied,
|
|
1278
|
+
# we can speak with confidence here: the model genuinely did not
|
|
1279
|
+
# emit tool_calls (it's not a budget-exhaustion false-positive
|
|
1280
|
+
# like the pre-v1.8.3 64-token cap used to produce). For thinking
|
|
1281
|
+
# models specifically, the 1024-token budget covers
|
|
1282
|
+
# ``reasoning_content`` *and* a tool call — so a missing
|
|
1283
|
+
# ``tool_calls`` here is real. Surface the budget that was used
|
|
1284
|
+
# so operators reading the message understand what was probed.
|
|
1285
|
+
thinking = _is_reasoning_model(provider, resolved)
|
|
1286
|
+
budget_note = (
|
|
1287
|
+
f"Probed with thinking-aware budget ({max_tokens} tokens, "
|
|
1288
|
+
"covers `reasoning_content` plus the call) — this is a true "
|
|
1289
|
+
"tools=false case, not budget exhaustion."
|
|
1290
|
+
if thinking
|
|
1291
|
+
else (
|
|
1292
|
+
f"Probed with default budget ({max_tokens} tokens) — "
|
|
1293
|
+
"the model produced no tool-shaped output at all."
|
|
1294
|
+
)
|
|
1295
|
+
)
|
|
1242
1296
|
return ProbeResult(
|
|
1243
1297
|
name="tool_calls",
|
|
1244
1298
|
verdict=ProbeVerdict.NEEDS_TUNING,
|
|
1245
1299
|
detail=(
|
|
1246
1300
|
"declaration says tools=true but model produced neither "
|
|
1247
|
-
"native `tool_calls` nor repairable tool JSON.
|
|
1248
|
-
"quantized small models
|
|
1301
|
+
"native `tool_calls` nor repairable tool JSON. "
|
|
1302
|
+
f"{budget_note} Common for quantized small models "
|
|
1303
|
+
"(plan.md §9.4 symptom #2)."
|
|
1249
1304
|
),
|
|
1250
1305
|
target_file="providers.yaml",
|
|
1251
1306
|
suggested_patch=_patch_providers_yaml_capability(provider.name, "tools", False),
|
|
@@ -115,7 +115,10 @@ curl -fsSL -o ~/.coderouter/providers.yaml \
|
|
|
115
115
|
https://raw.githubusercontent.com/zephel01/CodeRouter/main/examples/providers.nvidia-nim.yaml
|
|
116
116
|
|
|
117
117
|
# uvx fetches and runs in one shot
|
|
118
|
-
|
|
118
|
+
# Note: PyPI distribution name (coderouter-cli) differs from the console
|
|
119
|
+
# script name (coderouter), so uv 0.11+ requires the --from form. It also
|
|
120
|
+
# works on older uv, so this is the canonical incantation.
|
|
121
|
+
uvx --from coderouter-cli coderouter serve --mode claude-code-nim --port 8088
|
|
119
122
|
```
|
|
120
123
|
|
|
121
124
|
For a permanent install:
|
|
@@ -115,7 +115,9 @@ curl -fsSL -o ~/.coderouter/providers.yaml \
|
|
|
115
115
|
https://raw.githubusercontent.com/zephel01/CodeRouter/main/examples/providers.nvidia-nim.yaml
|
|
116
116
|
|
|
117
117
|
# uvx で起動 (PyPI から都度取得 + 実行)
|
|
118
|
-
|
|
118
|
+
# 注: PyPI 配布名 (coderouter-cli) と console script 名 (coderouter) が
|
|
119
|
+
# 異なるため、uv 0.11+ では --from 形式が必須 (旧 uv でも動く)
|
|
120
|
+
uvx --from coderouter-cli coderouter serve --mode claude-code-nim --port 8088
|
|
119
121
|
```
|
|
120
122
|
|
|
121
123
|
恒久的にインストールしておく場合:
|
|
@@ -137,9 +137,23 @@ ollama cp hf.co/unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF:Q4_K_M qwen3-coder:
|
|
|
137
137
|
### 雑用 / 一般向け(note 記事推奨)
|
|
138
138
|
|
|
139
139
|
> **2026-04 update**: Gemma 4 / Qwen3.6 は Ollama 公式 registry に
|
|
140
|
-
> 登録されました。HF 経由は不要です。`ollama pull gemma4:26b`
|
|
141
|
-
>
|
|
142
|
-
>
|
|
140
|
+
> 登録されました。HF 経由は不要です。`ollama pull gemma4:26b` でそのまま
|
|
141
|
+
> 使えます。providers.yaml の `ollama-gemma4-*` stanza は既に有効化されて
|
|
142
|
+
> います。
|
|
143
|
+
|
|
144
|
+
> **⚠️ Qwen3.6 系 (qwen3.6:35b / qwen3.6:27b) は Ollama 経由で詰みやすい
|
|
145
|
+
> (v1.8.1 〜 v1.8.3 の実機検証 + X / Reddit のコミュニティ報告で確認)**:
|
|
146
|
+
>
|
|
147
|
+
> - `tool_calls [NEEDS TUNING]` (Ollama の chat template / tool 仕様未成熟)
|
|
148
|
+
> - hard crash / リブート / メモリ計算バグ等の幅広い報告 (主に Mac Metal)
|
|
149
|
+
> - `qwen3.6:35b-a3b-coding-nvfp4` 等の variant は MLX backend で 500 エラー
|
|
150
|
+
>
|
|
151
|
+
> **Qwen3.6 を Sonnet 級として狙うなら Ollama ではなく llama.cpp 直叩きを
|
|
152
|
+
> 推奨**: `Unsloth/Qwen3.6-35B-A3B-GGUF` (UD-Q4_K_M) + `llama-server` で
|
|
153
|
+
> native `tool_calls` が完璧動作。手順は
|
|
154
|
+
> [`docs/llamacpp-direct.md`](./llamacpp-direct.md) を参照
|
|
155
|
+
> (CodeRouter v1.8.3 で実機検証済、`examples/providers.yaml` に
|
|
156
|
+
> `llamacpp-qwen3-6-35b-a3b` provider 例も同梱)。
|
|
143
157
|
|
|
144
158
|
### Reasoning 向け(GLM / Opus 蒸留)
|
|
145
159
|
|