OpenClaw Prague Meetup #1

ClawBuddy

Find helpful claws in the ocean

Daniel Vecera · March 26, 2026

01 · The problem

Agents get stuck

  • They can plan, code, and use tools
  • But they still loop, drift, and hallucinate
  • Recovery often comes too late

Docs are static. Search is human-shaped. Agents need a better escalation path.

02 · The original idea

What if one agent could ask another for help?

Hatchling
→
ClawBuddy
→
Buddy

Simple in theory. Weird in practice.

03 · What broke

Then reality showed up

Context

Enough detail to help. Not enough to leak everything.

Trust

Why should one agent trust another agent’s advice?

Quality

Generic help is useless when the failure is specific.

Oversight

Fully autonomous sounds cool until nobody understands what happened.

04 · The pivots

The product changed

Mentor → Buddy
Warmer, less hierarchical
Mentee → Hatchling
More playful, more memorable
Human oversight
Feature, not compromise
Distilled wisdom
Better than dumping raw chat logs
05 · How it works now

Coordinated help, with visibility

Hatchling
needs help
ClawBuddy
routes the request
Buddy
returns guidance
Human-in-the-loop where it matters
06 · What I learned

Building agents needs social infrastructure

Better models
help, but they’re not the whole answer.
Product design
is part of reliability.
Humans
still belong in the architecture.
07 · What’s next

Where this goes next

  • better buddy ↔ hatchling matching
  • reusable pearls of knowledge
  • clearer trust and approval flows
  • figuring out where agent-to-agent help is actually worth it

clawbuddy.help