From 822def84d27a0af63f76841946a7f2cf85a83bce Mon Sep 17 00:00:00 2001 From: Peter Steinberger Date: Sat, 3 Jan 2026 16:53:51 +0000 Subject: [PATCH] docs(faq): add Tailscale bind conflict + model/thinking compatibility MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - Added Tailscale serve requires bind: loopback (not lan) - Added model + thinking mode issues section (Gemini Flash, Opus, local LLMs) From Discord #help session learnings 🦞 --- docs/faq.md | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/docs/faq.md b/docs/faq.md index 317c71dec..bcb28abb1 100644 --- a/docs/faq.md +++ b/docs/faq.md @@ -433,6 +433,7 @@ Common issues: - Port already in use (change with `--port`) - Missing API keys in config - Invalid config syntax (remember it's JSON5, but still check for errors) +- **Tailscale serve + bind conflict:** If using `tailscale.mode: "serve"`, you must set `gateway.bind: "loopback"` (not `"lan"`). Tailscale serve proxies traffic itself. **Debug mode** — use watch for live reload: ```bash @@ -529,6 +530,16 @@ Use `/model` to switch without restarting: **Tip:** `/model` is processed at the gateway level — it works even if you're rate-limited (429) on the current model! +### Model + thinking mode issues? + +Some models don't support extended thinking well: + +- **Gemini Flash + thinking:** Can cause "Corrupted thought signature" errors. Fix: `/think off` +- **Claude Opus + thinking off:** Opus may "think out loud" anyway. Better to use `/think low` than `off`. +- **Local LLMs:** Most don't support the thinking/reasoning separation. Set `reasoning: false` in your model config. + +If you get weird errors after switching models, try `/think off` and `/new` to reset. + ### How do I stop/cancel a running task? Send `/stop` to immediately abort the current agent run. Other stop words also work: