OpenClaw 4.20 Update: SOUL.md, Kimi K2.6 and iMessage Finally Done Right

The OpenClaw 4.20 update lands and one change stands out for anyone building agents that talk to humans.

SOUL.md and IDENTITY.md actually work on GPT-5 and Codex now.

Your agent sounds like the character you wrote.

Not a corporate support bot.

Let me walk through that fix and everything else in the 4.20 release that matters for your business.

Why Personality Matters in the OpenClaw 4.20 Update

OpenClaw has two files called SOUL.md and IDENTITY.md.

These are where you tell your agent who it is — its tone, its style, its quirks.

Before 4.20, the newer OpenAI models like GPT-5 and Codex were kind of stiff about these files.

They'd read them.

But replies still felt like a corporate support robot.

What changed in 4.20

They made GPT-5 and Codex pick up that personality way better.

Your agent actually sounds like the character you wrote.

The OpenClaw team said it should feel "more like your weirdly capable little friend and less like a polished customer support automation".

Those are their exact words.

Why this matters for your business

If your agent handles DMs, messages, texts, or leads — it has to sound like you.

Not a generic chatbot.

Customers can smell a bot reply from ten feet away.

They switch off.

Your open rates drop.

Your conversion rates drop.

The 4.20 update makes that way easier to get right on GPT-5 and Codex.

What to put in your SOUL.md

Short list:

Tone (casual, direct, funny, straight, whatever fits you)
Catchphrases you actually use
Topics you care about
Topics you refuse to answer
How you open a reply
How you sign off

Keep it tight.

Don't try to describe your whole personality in 10,000 words.

A focused half-page beats a sprawling essay.

🔥 Want to build an agent that sounds like you, not a bot? Inside the AI Profit Boardroom, I've got a full 30-day OpenClaw roadmap that includes the exact SOUL.md structure I use. Four weekly coaching calls where you bring your setup and we fix it live. 2,800 business owners inside. Prompt library, member map, step-by-step videos. → Get the full roadmap here

Kimi K2.6 Is Now Bundled in the OpenClaw 4.20 Update

Before 4.20, using Moonshot's Kimi inside OpenClaw put you on K2.5.

Now it's K2.6 by default.

K2.5 still available if you need it.

Why K2.6 beats K2.5

Kimi is open-source.

Fast.

Cheap to run.

K2.6 is much better at thinking, replying, and handling tools.

Handling tools is the key improvement for agent workflows.

If you were already on Kimi, your agent just got smarter automatically.

Where K2.6 shows up

Web search uses K2.6 when you pick the bundled Kimi setup.

Media understanding uses it.

Token cost tracker knows K2.6 pricing so you see task cost before running.

Thinking keep all mode

New mode where the model thinks before every reply.

Doesn't guess.

Works out the answer first.

Kimi is one of the cheapest models you can run through OpenClaw.

Now it reasons like the expensive frontier models.

Lower cost + better answers.

iMessage Agents Actually Send in the OpenClaw 4.20 Update

This is the biggest practical fix.

Mac + newest macOS + AI agent through iMessage using Blue Bubbles?

Before 4.20, pain.

Messages wouldn't send.

Or they'd take a full minute and break.

Sometimes vanished into nothing.

The three fixes

Send time limit pushed from 10 seconds to 30 seconds.

System uses a private path on new macOS so text actually goes out.

Weird error breaking plain text on TAH removed.

What this opens up

Run an AI agent from your Mac.

Text customers over iMessage like a real person.

Messages actually land.

Local shop owner gets a text at 10pm about opening hours — agent replies straight away.

Coach gets a DM on weekend — agent books the lead into a call.

Tapback fallback

Agent reacts with a non-standard emoji.

Before 4.20, whole reaction failed.

Now OpenClaw falls back to a related standard reaction.

Customer still feels seen.

iMessage picked over SMS

Same number has both iMessage and SMS?

OpenClaw now picks iMessage first.

Before it sometimes sent SMS when the blue bubble was working.

Now it does the right thing every time.

Cron Jobs Rebuilt in the OpenClaw 4.20 Update

Scheduled tasks got a full rebuild.

What was broken

Jobs would say "delivered" but silently skip sending.

Jobs set to no-delivery would throw false errors.

Jobs running every hour piled up and ate memory.

Eventually the gateway crashed.

What got fixed

Runtime state split into its own file so your job list stays clean.

Main session delivery fixed — scheduled jobs land in the right chat.

Multi-channel jobs validated at save time so broken ones get caught before they run.

Example workflow

Local gym.

Agent texts every new sign-up on day 3 with a pep talk.

Agent texts again on day 7 asking how the first week went.

Set it up once in 4.20.

Runs forever.

Real errors when something breaks.

Not fake success.

Same idea works for coaches — day one welcome, day three check-in, day seven review call booking.

All on autopilot.

/think Command Smarter in the OpenClaw 4.20 Update

Before 4.20, telling a model to think when it couldn't would break mid-task.

Now OpenClaw checks first.

Knows what each model handles.

No random errors.

Thinking off stays off.

Some models used to quietly keep thinking.

Gone.

Agent is more reliable.

Security Locked Down in the OpenClaw 4.20 Update

Sneaky attacks blocked.

SSRF guard

Bad requests can't trick your server into doing something silly.

Workspace ENV injection blocked

Weird file dropped in your project folder can't silently change setup.

Agent can't touch its own config

Before 4.20, the AI could technically rewrite parts of its own config — change permissions, change tools.

Scary.

In 4.20 the model can't touch sandbox settings, trust rules, or MCP server configs.

Tries a sneaky edit? Blocked.

Device pairing tightened

Device connected to your OpenClaw sees only its own pairing info.

Can't peek at or approve other devices.

Failed connection returns a real error, not a generic one.

Memory and Sessions in the OpenClaw 4.20 Update

Clean-up work under the hood.

Pile-up fixed

Old sessions used to accumulate until the gateway ran out of memory.

In 4.20 there's a cap and age-based cleanup.

Sessions get pruned automatically.

Gateway healthy at hundreds of jobs per day.

Cost tracking honest

Session saves used to sometimes count costs 2x or 10x.

Now snapshots properly.

Month-end number is accurate.

/new and /reset clear state

Before, fresh sessions kept stale model choices.

You'd think you were on Kimi K2.6 but stuck on a fallback from three mistakes ago.

Now they reset cleanly while keeping intentional choices.

Video notes + links to the tools 👉 here

Channel Updates in the OpenClaw 4.20 Update

Quick run-through.

Auto-reply context-aware

DM: helpful reply.

Group: quiet unless tagged.

No more group chat spam.

Setup wizard cleaner

Proper walkthrough with loading spinner, clear headings, clear warnings.

Anyone can follow it.

Telegram polling grace period

Longer before OpenClaw thinks the connection's broken.

Stateless reactions work properly.

Matrix allow-list without restart

Change settings on the fly without restarting the channel.

Manus live draft preview

Agent shows a live draft as it writes.

Final reply lands in place.

Feels natural.

Discord fixes

/think only shows valid options for your model.

Missing slash commands don't crash.

What the OpenClaw 4.20 Update Means for Your Business

Five takeaways:

1. Kimi K2.6 inside: smarter, cheaper default model.

2. iMessage agents real: huge channel for customer-facing businesses.

3. Scheduled tasks reliable: cron finally works.

4. Security locked down: team handoff without fear.

5. Personality picks up on GPT-5 and Codex: agents sound like you.

If you pair this with other local-model setups, my Ollama + Hermes guide is a good companion.

🔥 Want to make the most of the OpenClaw 4.20 update for your business? The AI Profit Boardroom has the full OpenClaw training — install, wire, automate, secure. Weekly coaching calls, 2,800 members, prompt library, 30-day roadmap, member map. → Join the Boardroom

Honest Take on the OpenClaw 4.20 Update

Most AI agent tools are rough.

They break.

They hallucinate.

They run up weird numbers.

OpenClaw 4.20 is still going to be buggy in spots.

I've got to admit that.

When I've checked Hermes, it sometimes feels smoother.

Both tools are improving.

I switch between them depending on the task.

Sometimes an update breaks your setup and you have to restart from scratch.

Back up OpenClaw before you install 4.20.

If you want to compare the two, my Hermes agent workspace post goes deep on the Hermes side.

How to Install the OpenClaw 4.20 Update

Type update inside OpenClaw.

Pulls the latest.

Under a minute on most setups.

OpenClaw 4.20 Update FAQ

Will the OpenClaw 4.20 update improve my agent's voice?

Yes — GPT-5 and Codex pick up SOUL.md and IDENTITY.md much better now.

Does the OpenClaw 4.20 update work with custom SOUL.md files?

Yes — your existing files get picked up automatically after update.

Might be worth tightening them to take advantage of the better parsing.

Is Kimi K2.6 really free?

Yes — open source, bundled with OpenClaw.

Do I need to reinstall Blue Bubbles after the OpenClaw 4.20 update?

No — the fix is inside OpenClaw, Blue Bubbles stays the same.

Will the OpenClaw 4.20 update break anything?

Sometimes updates break setups.

Back up first.

How long does the OpenClaw 4.20 update take to install?