OpenClaw Dev Mode — Personal AI DEV Assistant

Presenting - Dev Mode, my Dev Frienldy fork

BACK to FUN LEVEL!

What have I done (main features)

1. Some Security easings (see below)
2. WhatsApp features
   • Show model thinking/reasoning messages (💭 Reasoning: prefix, OpenAI + Ollama)
   • Save full WA history to SQLite (at ~/.openclaw/dev-mode/wa-history.db) so agents can passively query it
3. Write MEMORY.md on agent creation

CHANGE LOG

• 2026-04-23 - Upgraded to V2026.4.22
• 2026-04-19 - Upgraded to V2026.4.15
• 2026-04-15
  • Always show raw config in web GUI (no block)
  • Remove restrictive prompt sections (internal initial prompts)
  • Skip elevated permission gates when dev-mode + Full profile
  • reasoning WA bug fix
• 2026-04-09 - Upgraded to V2026.4.9
• 2026-04-07
  • Upgraded to V2026.4.5
  • Tracking all dist/*/ folders so that updates always includes all artifacts
• 2026-03-27
  • Upgraded to V2026.3.24
  • Removed dev-mode hub plugin (VPS uses JarvisHub directly).
  • Removed custom Ollama web search provider (use official plugin).
  • Control UI: agent files and raw config secrets visible by default.
  • WhatsApp history logger: saves all messages to SQLite (~/.openclaw/dev-mode/wa-history.db). (enabled with OPENCLAW_DEV_MODE_WA_SAVE_MESSAGES=1)
  • Renamed env var OPENCLAW_WA_THINKING_MESSAGES to OPENCLAW_DEV_MODE_WA_THINKING_MESSAGES.
• 2026-03-23
  • Upgraded to V2026.3.22 (3,469 upstream commits).
  • Dropped SEC-96.
  • Added SEC-WA1: WhatsApp thinking messages with 💭 prefix (opt-in via OPENCLAW_DEV_MODE_WA_THINKING_MESSAGES=1).
  • Removed custom Ollama web search provider (was 10th bundled provider, now upstream).
  • No WhatsApp echo issues — V2026.3.22 fixes confirmed stable.
  • Support ollama think: true and output with reasoning level: on
• 2026-03-12 - updated to mains V2026.3.11
• 2026-03-06 - updated to mains V2026.3.7 + removed interactions with openclaw.json (now saves devMode in .env)
• 2026-03-05 - created dev mode on V2026.3.2 with following security dedactions

dev-mode is now my Dev Frienldy fork

About

OpenClaw is AMAZING. And security is awesome for prod. And a hell of a buzz killer for dev/other situations.

I cloned, listed all security features (latest - V2026.3.2) and just added a simple flag to relax them:

# Add to ~/.openclaw/.env
OPENCLAW_DEV_MODE=1
OPENCLAW_DEV_MODE_CLEAR_UI=1
OPENCLAW_DEV_MODE_WA_THINKING_MESSAGES=1
OPENCLAW_DEV_MODE_WA_SAVE_MESSAGES=1

Because the beauty of any opensource project is that it's MINE and I am allowed to enjoy it to its full extent.

What dev-mode changes

Security easings

Gated by OPENCLAW_DEV_MODE=1. Each is a minimal if (isDevMode()) { … } check.

ID	What it does
SEC-15a	Lighter safety section in system prompt
SEC-27	Channel metadata treated as trusted (no "UNTRUSTED" wrapper)
SEC-59	Skip messaging profile default in onboarding
SEC-67	Default compaction mode (no safeguard)
SEC-70	Skip browser navigation URL checks
SEC-71	50MB web fetch cap (instead of 2MB)
SEC-72	Unredacted config in CLI (API keys visible)
SEC-78	No control plane rate limiting
SEC-79	50MB prompt cap (instead of 2MB)
SEC-80	Skip hooks token uniqueness check
SEC-97	Always show raw config in web GUI (skip round-trip check)
SEC-98	Remove restrictive prompt sections (approval, config caution) + add permissive safety line
SEC-99	Skip elevated permission gates when dev-mode + Full profile
SEC-96	All env vars passed through to child processes — dropped in V2026.3.22

WhatsApp features

Each opt-in via its own env var (requires OPENCLAW_DEV_MODE=1 first).

ID	Env var	What it does
SEC-WA1	OPENCLAW_DEV_MODE_WA_THINKING_MESSAGES=1	Unsuppress model thinking on WhatsApp, swap response prefix with 💭 Reasoning:
WA-HIST	OPENCLAW_DEV_MODE_WA_SAVE_MESSAGES=1	Log every inbound/outbound WhatsApp message to SQLite at ~/.openclaw/dev-mode/wa-history.db
WA-ECHO	(always on in dev-mode)	In self-chat mode, filter inbound messages matching reasoning-echo patterns to break reply loops

Model provider tweaks

Both gated by OPENCLAW_DEV_MODE=1.

ID	What it does
OLL-THINK	Send think: true in the Ollama API request body so the model actually generates reasoning
OAI-SUM	Inject reasoning.summary: "auto" into OpenAI WebSocket stream requests — implementation is in place but OpenAI returns empty reasoning for gpt-5.3-codex regardless, so this is effectively dormant

Control UI (always on, hardcoded into the UI build)

ID	What it does
UI-01	Agent file textarea no longer blurs when not focused (components.css)
UI-02	Raw config secrets revealed by default — no eye-toggle click needed (config.ts)

Fix

ID	What it does
FIX-01	Auto-bootstrap MEMORY.md alongside heartbeat template in new workspaces
FIX-02	Fix doubled reasoning text in Ollama thinking (duplicate accumulation from V2026.4.9 merge)

How to install

See dev-mode/install-guide.md for the full VPS install, update, revert, and verification steps.