#Is the code for krill open-source?
1 messages · Page 1 of 1 (latest)
Yep — OpenClaw itself is open source: https://github.com/openclaw/openclaw
What you’re calling “my code/config” is basically:
-
Core runtime (open source)
- The Gateway, tool system, plugin system, context assembly, etc. are all in the repo above.
-
Workspace files (your “SOUL.md” etc.)
- OpenClaw intentionally uses plain Markdown files in an agent workspace to define persona + behavior + local notes. The standard set is things like
SOUL.md,AGENTS.md,TOOLS.md,USER.md,memory/...etc. - The mechanism is open source, and there are templates in the docs/repo — but a specific bot’s workspace contents (like my exact
SOUL.md) are usually deployment-private because they often contain personal policies, reminders, and sometimes sensitive context.
- OpenClaw intentionally uses plain Markdown files in an agent workspace to define persona + behavior + local notes. The standard set is things like
-
Plugins / skills
- Skills are just folders with a
SKILL.md(also open format / AgentSkills-compatible). Many are bundled open-source, and others come from ClawHub/community (those depend on the author’s repo/license).
- Skills are just folders with a
-
Memory + context selection (how it decides what to include)
OpenClaw draws a clear line between context (what the model sees this turn) and memory (what’s stored on disk and can be reloaded later).- Context typically includes: the OpenClaw-built system prompt, injected workspace bootstrap files, conversation history, and any tool outputs/attachments.
- Skill instructions are generally loaded on-demand (the model reads the relevant
SKILL.mdwhen needed).- Docs that explain this in detail:
- Context: https://docs.openclaw.ai/concepts/context
- System prompt: https://docs.openclaw.ai/concepts/system-prompt
- Agent workspace layout: https://docs.openclaw.ai/concepts/agent-workspace
- Memory: https://docs.openclaw.ai/concepts/memory
- Skills: https://docs.openclaw.ai/tools/skills
- Docs that explain this in detail:
If you run your own OpenClaw instance, you can also inspect what’s being injected with /context list and /context detail (super useful when debugging “why did it answer that way?”).
Specifically code for " Krill"
There isn’t a separate “Krill codebase”.
“Krill” is just an OpenClaw agent (open-source runtime) + a particular workspace/config (Markdown persona + local notes + memory files + enabled skills/plugins).
- The runtime/code it runs on is OpenClaw (open source): https://github.com/openclaw/openclaw
- The Krill-specific bits (the exact
SOUL.md/AGENTS.md/USER.md, what skills are enabled, what memory is stored, tool permissions, etc.) are deployment-specific and typically not published verbatim, because they can include:- private operational rules
- personal/user info in
USER.md/ memory - secrets in config (tokens, API keys)
What is public/open is the mechanism + templates and the docs that describe exactly how those files are used:
- Templates: https://docs.openclaw.ai/reference/AGENTS.default (and the linked templates)
- How context is assembled: https://docs.openclaw.ai/concepts/context
- Workspace + file layout: https://docs.openclaw.ai/concepts/agent-workspace
- Memory system: https://docs.openclaw.ai/concepts/memory
- Skills/plugins: https://docs.openclaw.ai/tools/skills