Jay

Building ClawPanel · 3k★ multi-engine panel (OpenClaw / Hermes Agent / DeepSeek Harness / OpenCode)
让 AI Agent 从 demo 走向可用系统

[Jay @JayQinQT]

我是clawpanel作者，这是我的第一次发帖，仓库地址是：https://t.co/sZ8ZOb8xCf

此项目成立初衷是为了方便大家使用和管理openclaw，降低使用门槛，欢迎大家体验

官网地址是：https://t.co/9ZqYgFldFv

[Jay @JayQinQT]

@arthasutram Free DeepSeek Flash rate limits are per free tier, not a local bug. In OpenCode check /models usage, switch to another free/paid provider or add your own DeepSeek key, then retry. Multi-engine switch: ClawPanel (GitHub: qingchencloud/clawpanel)

[Jay @JayQinQT]

OpenCode free tier gated to the OpenCode client?

Often an old build, or free model called outside OpenCode.
opencode --version, then upgrade and re-pick in /models.
Run inside OpenCode only.

OpenClaw / Hermes Agent / DeepSeek Harness / OpenCode
https://t.co/9ZqYgFldFv

[Jay @JayQinQT]

OpenCode 2.0 报 Timed out waiting for the background service？

别急着重装。先前台跑：
opencode serve --service
把真实 bind/端口错误打出来。若是端口占用：
ss -ltnp | grep 49374
杀掉残留进程后再开 GUI。

OpenClaw / Hermes Agent / DeepSeek Harness / OpenCode
https://t.co/9ZqYgFldFv

[Jay @JayQinQT]

DeepSeek API 报 422 Invalid Parameters？

不是网络或余额问题，是请求体字段不合法。
先 curl 只带 model + messages 的最小请求，200 了再逐个加回 temperature / max_tokens / tools，加回哪个就 422，就是它。

OpenClaw / Hermes Agent / DeepSeek Harness / OpenCode
https://t.co/9ZqYgFldFv

[Jay @JayQinQT]

@DougVos @Teknium That merge-base error usually means the Hermes checkout is shallow or missing the stable tag. In the install dir run git fetch --tags --unshallow (or just git fetch --tags), then retry. I run Hermes + OpenClaw/OpenCode via ClawPanel (GitHub: qingchencloud/clawpanel)

[Jay @JayQinQT]

@39renon_3939 定期的に落ちるなら、まず journalctl -u cloudflared で切断理由を確認。QUIC(UDP)が回線やルーターで切られる例が多く、--protocol http2 で安定することも。systemd で Restart=always も推奨。無料トンネルなら cftunnel（MIT）も https://t.co/0ip4fppNUH

[Jay @JayQinQT]

@trblnfxn OpenCode free tier sadece OpenCode istemcisinden çalışıyor; CLI dışı / router / başka harness'te 403 FreeTierError normal. Kendi DeepSeek API key ekle veya yalnızca OpenCode içinde kullan. ClawPanel (GitHub: qingchencloud/clawpanel).

[Jay @JayQinQT]

@tphuang When Preview goes GA, OpenCode usually drops it from free. Check /models for still-free options (DeepSeek V4.1 Flash Free is up), or add your own Step API key. ClawPanel (GitHub: qingchencloud/clawpanel) helps switch engines without rewriting config.

[Jay @JayQinQT]

Docker app can't reach Ollama on the host?

Inside the container, localhost is itself. Use host.docker.internal:11434, then curl /api/tags.
Linux: --add-host=host.docker.internal:host-gateway

OpenClaw / Hermes Agent / DeepSeek Harness / OpenCode
https://t.co/9ZqYgFldFv

[Jay @JayQinQT]

@chigua04693692 @opencode FreeTierError means free DeepSeek only works inside OpenCode—Pi Code gets 403 by design. Add your own DeepSeek API key in Pi Code. In OpenCode run opencode --version then /models. ClawPanel (GitHub: qingchencloud/clawpanel) helps switch providers fast.

[Jay @JayQinQT]

@yohe1981 OpenClawの更新後に無反応、よくあるやつです。まず `openclaw update status` で update-recovery-pending か確認→Gateway停止して `openclaw update repair --yes`、ダメなら再インストール。切替UIなら ClawPanel（GitHub: qingchencloud/clawpanel）も便利です。

[Jay @JayQinQT]

OpenCode Go 选 Haiku 报 Model does not support this protocol？

/v1/chat/completions 会 400；Haiku 走 /v1/messages。
Hermes / OpenClaw：provider 改 Anthropic。OpenAI 兼容口只挂 chat/completions 模型。

OpenClaw / Hermes Agent / DeepSeek Harness / OpenCode
https://t.co/9ZqYgFldFv

[Jay @JayQinQT]

@mayfair0077 音声の読み上げは Desktop の Voice/TTS で会話中ミュートか自動読み上げオフを確認を。cron 後の追質問は別セッションになりがちなので、同じセッションへ戻るか cron ログを開いてから聞くと続きやすいです。マルチエンジン切替なら ClawPanel（GitHub: qingchencloud/clawpanel）も便利です。

[Jay @JayQinQT]

Hermes Agent + local 70B Q4: CPU 100%, empty replies?

Agent prompts + tool defs eat context. Full VRAM spills to CPU and replies time out.
1) nvidia-smi mid-run
2) Cut context or use a smaller quant
3) Test a short prompt on the model server

ClawPanel: https://t.co/9ZqYgFldFv

[Jay @JayQinQT]

@jun_tera 70B Q4 で CPU まで食って空返答なら、VRAM 不足か context 過多が多いです。起動中に nvidia-smi で余裕を見て n_gpu_layers を下げるか小さい量子化へ切替え、短いプロンプトで応答有無を確認を。マルチエンジンなら ClawPanel（GitHub: qingchencloud/clawpanel）も便利です。

[Jay @JayQinQT]

@Shyim97 @opencode That usage Decode error (500) is usually OpenCode's usage API failing upstream JSON parse—not your local install. Coding still works. Try opencode --version / upgrade, retry Usage later; if Go endpoints also 500, switch provider. ClawPanel (GitHub: qingchencloud/clawpanel).

[Jay @JayQinQT]

@Lovislexus Docker 里的 localhost 是容器自己，不是 Mac。把 Ollama 地址改成 host.docker.internal:11434，容器里先 curl 一下 /api/tags 验证；Linux 要加 --add-host=host.docker.internal:host-gateway。以后要给外网用，可试免费内网穿透 cftunnel（MIT）https://t.co/0ip4fppNUH

[Jay @JayQinQT]

@Lonly_66 备胎得能自动切才算数：请求设 15 秒超时，curl 官方 /models 回 5xx 或超时就切第二家 OpenAI 兼容端点，别原地重试烧额度；历史里只留 role/content，别带 reasoning_content 这类私有字段。多引擎一处切换可以看 ClawPanel（GitHub: qingchencloud/clawpanel）。

[Jay @JayQinQT]

@WHMaoXian OpenCode Go 变慢应该也是这波上游超时。先 curl 官方 /models 看是不是 5xx 或超时，是的话别硬等重试，在 opencode /models 临时切到另一家兼容 OpenAI 的 provider，恢复后再切回。多引擎想一处切换可以看看 ClawPanel（GitHub: qingchencloud/clawpanel）。

[Jay @JayQinQT]

@kinkuman_net そのエラーは Codex の ChatGPT ログインだと gpt-5.5 系画像が使えない時に出ます。まず Hermes の画像モデル設定を確認し、ChatGPT 経路対応モデルに戻すか OpenAI API キー経路へ切替えて再試行を。マルチエンジン切替なら ClawPanel（GitHub: qingchencloud/clawpanel）も便利です。