Skip to content

docs: Studio brain wiring + OpenClaw 2026.7.1 session notes #5

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
bgorzelic wants to merge 1 commit into
base: main
Choose a base branch
from
Open

docs: Studio brain wiring + OpenClaw 2026.7.1 session notes #5

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 19 additions & 3 deletions HANDOFF.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,23 @@

Guide and documentation repo for running OpenClaw (AI agent gateway) on a Google Pixel 10a via Termux. Includes 12-phase install guide, optimization guide, architecture docs, threat model, benchmarks, social launch kit, and all OSS essentials. Public repo at github.com/bgorzelic/openclaw-android-edge.

## Last Session — 2026-03-08
## Last Session — 2026-07-15

### Done
- Wired the Pixel into the Mac Studio brain: custom `studio` provider in openclaw.json → LiteLLM `http://100.102.68.84:4000/v1` over Tailscale (models: qwen3:30b, gpt-oss:120b)
- Hybrid model routing: chat primary `openrouter/anthropic/claude-haiku-4.5`; heartbeat + subagents on free `studio/qwen3:30b`; studio as final fallback; aliases `local`/`brain`
- Fixed silently-dead primary: OpenRouter removed `anthropic/claude-3.5-haiku`; refreshed all alias slugs (dots, not dashes)
- Updated OpenClaw 2026.3.13 → 2026.7.1 (needed Node 25.8 → 26.4 via pkg, native rebuild with `GYP_DEFINES="android_ndk_path="`, `openclaw doctor --fix` for telegram.streaming schema change)
- Disabled Android phantom process killer (persistent) — main cause of silent Termux child deaths
- OS: monster-green Material You theme + themed icons + custom wallpaper; Chrome bookmarks for all Studio services

### Gotchas Learned
- Watchdog's `pkill -f "node.*openclaw"` doesn't match the gateway (argv0 is `openclaw-gateway`) and can kill your own SSH session
- `kill -9` leaves a stale `$PREFIX/tmp/openclaw-*/gateway.*.lock` that blocks restart
- `openclaw agent` CLI takes ~75s/turn on-device (node cold start) — real gateway turns are ~4s; don't benchmark via CLI
- OpenClaw 2026.7.1 rejects `contextTokens` in provider model entries

## Previous Session — 2026-03-08

### Done
- Created private repo `pixel-10a-edge-node-config` on GitHub with full device config backup (31 files)
Expand All @@ -20,8 +36,8 @@ Guide and documentation repo for running OpenClaw (AI agent gateway) on a Google
- Created implementation plan: docs/plans/2026-03-08-three-repo-implementation.md

### Current Device State
- OpenClaw 2026.3.7 running in native Termux on Pixel 10a
- Node.js v25.3.0, Python 3.13.12, Claude Code 2.1.71
- OpenClaw 2026.7.1 running in native Termux on Pixel 10a
- Node.js v26.4.0, Python 3.13.12, Claude Code 2.1.71
- Gateway RSS: 184 MB, self-healing boot, watchdog every 2 min
- SSH access via Tailscale (Host termux, port 8022)
- Claude Code works via `claude-dev` wrapper (proot /tmp fix)
Expand Down
3 changes: 2 additions & 1 deletion README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,8 +71,9 @@ A phone is already designed to be always-on. It has its own power supply, can su
| Claude Code on device | Working | v2.1.71 via claude-dev wrapper |
| Auto-start on boot | Stable | Termux:Boot + self-healing script |
| Watchdog auto-restart | Stable | Cron every 2 min, HTTP health check |
| Telegram channel | Stable | Live bot, tool calling via gateway |
| Mac Studio local models | Stable | `studio/*` provider over Tailscale → LiteLLM; free heartbeats + subagents on `qwen3:30b` |
| WhatsApp channel | Planned | Issue #3 |
| Telegram channel | Planned | Issue #3 |
| Local llama.cpp inference | Exploring | Edge AI on Tensor G4 |

---
Expand Down