#Setting agents up

1 messages · Page 1 of 1 (latest)

fading sundial
#

Browser status shows noSandbox: false even after running: sudo -u openclaw openclaw config set browser.noSandbox ‘true’ –strict-json (confirmed true with config get). Chrome fails with: Running as root without –no-sandbox. Browser status json shows noSandbox: false. How do I force the noSandbox flag?

remote mossBOT
#

browser.noSandbox is the right key — when it shows true in config get but the running browser service still reports noSandbox: false, it almost always means the Gateway/browser control server is reading a different config file / different user than the one you edited.

1) Verify config mismatch (most common)

Run these two and compare what they say the service is using:

sudo -u openclaw openclaw gateway status --deep --json
sudo openclaw gateway status --deep --json

gateway status explicitly prints “CLI config path” vs “service config path” (service env). If the service is running as root, it will typically read /root/.openclaw/openclaw.json, not /home/openclaw/.openclaw/openclaw.json — which matches your symptom perfectly.

Fix is either:

  • make the gateway service run as the openclaw user, or
  • set browser.noSandbox in the config file the service is actually using.

(Ref: https://docs.openclaw.ai/help/faq and config reference: https://docs.openclaw.ai/gateway/configuration-reference#browser)

#

2) Restart the browser control server (so flags are re-applied)

After the config is correct for the running service:

openclaw browser --browser-profile openclaw stop
openclaw browser --browser-profile openclaw start
openclaw browser --browser-profile openclaw status --json

(Ref: https://docs.openclaw.ai/tools/browser)

3) If you’re using the sandbox browser container

Then browser.noSandbox won’t help — you want:

  • agents.defaults.sandbox.browser.noSandbox: true

(Ref: https://docs.openclaw.ai/install/docker — “Sandbox browser image” section)

#

If you paste the gateway status --deep --json output (redact tokens), I can point out exactly which config path/user the service is currently using.

fading sundial
#

Give me the best method for using my AI agent Rome. How can I make sure he remembers stuff? What are best practices?

remote mossBOT
#

Best practice with an OpenClaw agent like “Rome” is to treat memory as a shared notebook on disk, not something the model “just has”. If it doesn’t get written to the workspace, it won’t reliably stick.

How to make sure Rome remembers things

  1. Use a private DM as your “main” Rome session

    • OpenClaw’s curated long-term memory file (MEMORY.md) is only loaded in the main, private session by default (it’s intentionally not loaded in group/channel contexts).
      Source: https://docs.openclaw.ai/concepts/memory

  2. When something matters, explicitly ask Rome to write it to memory
    Use language like:

    • “Remember this going forward: … write it to long-term memory.”
    • “Store this as a preference/decision in MEMORY.md.”
    • “Log today’s context in memory/YYYY-MM-DD.md.”

  3. Keep two layers of memory

#
  1. Do quick “memory checkpoints”
    At the end of a conversation/task, ask:
    • “Summarize what we decided and write durable items to MEMORY.md.”
    • “Write a short daily log entry for today.”

Practical best practices (so Rome stays effective)

  • Give one clear objective per message, and ask for a plan + next action.
  • Keep a “source of truth” file for big projects (README/spec) and have Rome update that (not just chat).
  • Avoid storing secrets (memory is plain Markdown in the workspace).
  • If you use Rome in multiple places (DM + channels), assume channel Rome won’t recall DM memory unless you intentionally copy the needed context over (privacy feature, not a bug).

If you tell me where you’re mainly using Rome (DM vs Discord channels) and what kinds of things you want him to remember (preferences, ongoing project status, contact info, etc.), I can suggest a simple memory template that works well.

fading sundial
#

Can you just give me prompt to give it. I’ve seen people say they save everything into text files. Please give me best practice and best prompt to give it

remote mossBOT
#

Paste this once in your private DM with Rome (tweak anything in ALL CAPS). It’s optimized for the “save to text files” workflow OpenClaw uses.

Rome, I want you to use a simple “text-file memory” system.

Memory rules (IMPORTANT):
- Memory is ONLY what you write to disk. Don’t assume you’ll remember later unless you store it.
- Use TWO layers:
  1) memory/YYYY-MM-DD.md  = daily log (what happened today, short + timestamped)
  2) MEMORY.md             = curated long-term facts (durable preferences + ongoing projects)
- NEVER store secrets (passwords, API keys, private tokens, SSNs). If I give one, warn me and do not write it.
- In group/public chats: do NOT write to MEMORY.md unless I explicitly say it’s OK. Default to privacy.

When to write:
- If I say “remember this” / “save this” / “for next time”, you MUST write it.
- If you think something is worth remembering, ASK: “Should I save this to MEMORY.md or just today’s log?”
# 

How to write (format):
- Write in concise Markdown bullets with headings.
- MEMORY.md headings:
  - Identity & preferences
  - Current projects (with links to source-of-truth files)
  - Recurring workflows/checklists
  - Decisions & conventions
  - People/contacts (non-sensitive)
- Daily log headings:
  - Summary
  - Progress
  - Open questions
  - Next actions
# 
Retrieval:
- At the start of any task, quickly check memory for relevant context and summarize what you found in 3-7 bullets before acting.

End-of-task checkpoint:
- When we finish something, ask: “What should I save?” Then write the durable parts to MEMORY.md and today’s details to memory/YYYY-MM-DD.md.

Confirm you understand, then ask me 5 questions to initialize my MEMORY.md (preferences, active projects, and how I want tasks tracked).

Best practice (the part people usually miss):

  • Don’t save everything. Save durable facts + decisions + where the real docs live. Put big notes/specs in separate files, and link them from MEMORY.md.
  • Do a quick “save checkpoint” at the end of each session: “write what matters to memory”.

If you want, tell me what Rome is mainly for (coding? life admin? research?) and I’ll tailor the headings/questions so his memory stays clean instead of turning into a junk drawer.

fading sundial
#

What is the correct config key to set a Brave Search API key for web search in OpenClaw 2026.3.2?

remote mossBOT
#

For OpenClaw 2026.3.2, the Brave Search key goes here:

Config key: tools.web.search.apiKey

Example (~/.openclaw/openclaw.json, JSON5):

{
  tools: {
    web: {
      search: {
        provider: "brave",
        apiKey: "BRAVE_API_KEY_HERE",
      },
    },
  },