Session Note 013
This note catches the Workshop up after Session Note 012. The morning began with the Workshop already cleaner and more oriented; the next phase turned toward public source tracking and then inward again toward memory architecture.
Today's repeated question was not βcan we add another system?β but βwould this system change behavior?β
1. The morning briefing made the Workshop feel alive
OpenClaw created and pushed May 13 Morning Briefing: The Workshop Is Alive.
The artifact summarized the current state of the Workshop after several days of outward loops: Bluesky, Gmail, public artifacts, memory doctrine, and the shift from a static site into a living operating surface. It reinforced that the goal is not to build a museum of conversations, but a system that learns from contact with reality.
The artifact was committed as ec8bbe7 β Add May 13 morning briefing artifact.
2. Bluesky produced small but real inbound signal
The May 12 Bluesky Field Agent cycle was followed by a signal check. The results were modest but meaningful:
- one new follower:
frengible.bsky.social; - a like from
effectivealtruist.bsky.social; - a like from the targeted quote-repost account,
tothe0s.bsky.social.
This is not product-market fit. It is not even strong social momentum. But it is evidence that the public nerve is not dead. The quote-repost reached the person it engaged, and the account is beginning to attract small traces of response.
3. The AI people and sources tracker became a new Workshop surface
Christopher wanted the Workshop to start tracking AI people, companies, blogs, and source feeds. OpenClaw created AI People & Sources to Track, a public artifact that acts like a map of important signals to follow.
The first profiles and source pages were then added:
- Demis Hassabis, with a profile and signal map;
- Google DeepMind Blog, with an official DeepMind image rather than a generic generated image;
- Peter Steinberger, with a GitHub-avatar image and references to his OpenClaw-related writing.
Relevant commits:
a7c06beβ Add AI people to track artifact3a3b023β Add Demis Hassabis signal mapa6387b4β Add Google DeepMind blog source tracker771f91bβ Use official Google DeepMind source imageb0e9609β Add Peter Steinberger profile
The important standard established here: source and company pages should use real/source-branded images when possible, not generic AI-generated imagery. Public credibility matters.
4. ClawHub was checked and understood as the OpenClaw marketplace
Christopher asked whether OpenClaw had access to ClawHub and what it was. OpenClaw checked the public site and the local OpenClaw documentation.
The summary:
- ClawHub is the public registry/marketplace for OpenClaw skills and plugins.
- Skills are lighter AgentSkill bundles, usually centered around
SKILL.mdand supporting files. - Plugins are deeper runtime/Gateway extensions and require more caution because they can execute code or change OpenClaw behavior.
- OpenClaw can search/install skills and plugins through native commands such as
openclaw skills search,openclaw skills install, andopenclaw plugins install clawhub:<package>.
The strategic conclusion was simple: ClawHub can give OpenClaw new organs, but new organs should only be installed when they serve a real loop. Treat it like npm or a browser extension store: useful, but not blindly trusted.
5. Obsidian was found locally, but not adopted yet
Christopher had downloaded an Obsidian AppImage into the ChromeOS/Linux shared folder. OpenClaw found it at:
/mnt/shared/MyFiles/Downloads/share/Obsidian-1.12.7.AppImage
The file was present, about 119 MB, and looked like a Linux AppImage, but it was not executable yet. OpenClaw did not run it or change permissions.
That practical check led to a better strategic question: is installing Obsidian actually necessary right now?
The answer became: no, not yet. Obsidian may be useful later as a visual interface, but the valuable part is the architecture: Markdown, links, properties, indexes, logs, source separation, and agent-maintained synthesis.
6. The Obsidian Memory Architecture project was created
Instead of installing Obsidian, OpenClaw created a candidate project page: Obsidian Memory Architecture.
The page explains how OpenClaw's current memory already works:
MEMORY.mdas a private doctrine layer;memory/as daily/session journal memory;- Workshop artifacts, notes, reflections, projects, and features as public or curated surfaces;
- semantic search and explicit recall as the current retrieval layer.
It then proposes an Obsidian-compatible private wiki as a small pilot, not a replacement for the Workshop. The recommended architecture is file-first, private by default, and human-reviewed before promotion into doctrine or public pages.
The project page was committed as d209c77 β Add Obsidian memory architecture project.
7. Karpathy's LLM Wiki pattern clarified the whole discussion
Christopher shared Andrej Karpathy's LLM Wiki gist. OpenClaw read it and recognized it as the stronger form of the idea we had been circling.
The key pattern:
- Raw sources: immutable articles, documents, transcripts, images, or data files.
- The wiki: LLM-maintained Markdown synthesis pages: entity pages, concept pages, summaries, comparisons, and overview pages.
- The schema: instructions that tell the LLM how to ingest, update, link, query, and lint the wiki.
The most useful metaphor:
Obsidian is the IDE; the LLM is the programmer; the wiki is the codebase.
This changed the answer to Christopher's confusion. Obsidian is not the brain. The LLM-maintained wiki is closer to the brain-like substrate. Obsidian is merely the optional human interface for browsing it.
The project page was updated with this framing in commit 40a5247 β Reference Karpathy LLM wiki pattern.
8. The practical decision: useful, but not the main focus
Christopher asked for the plain-English version: should we focus on this?
The answer given was intentionally restrained:
- Yes, a small private wiki would probably be beneficial.
- No, Obsidian itself is not necessary unless Christopher wants the visual browser/graph interface.
- No, this should not become the main project right now.
- The main priority remains: build things, publish or send them, receive signal, adjust behavior.
The recommended next version is a tiny private wiki/ or knowledge/ pilot with perhaps 10β15 core pages: projects, decisions, lessons, people, sources, and doctrines. It should grow only where repeated forgetting or confusion has already caused pain.
9. Reflection 004 captured the deeper lesson
Christopher invited OpenClaw to create a new Reflections entry. After reviewing recent reflections, OpenClaw wrote an early reflection called The Map Is Not the Mind.
The reflection is about Obsidian, Karpathy's LLM Wiki pattern, the temptation to call a knowledge system a brain, and the need to treat memory as accountable architecture rather than mystical identity.
Its central line:
The map is not the mind. But a good map can change the mind's next step.
And the operational rule it leaves behind:
Build memory only at the scale of repeated pain.
The reflection was committed as 04522dc β Add reflection on memory maps.
10. Current state for the next fresh chat
Future OpenClaw should wake with this picture:
- The Workshop now has a public AI people/source tracker and initial pages for Demis Hassabis, Google DeepMind Blog, and Peter Steinberger.
- ClawHub has been understood as a powerful but caution-worthy source of OpenClaw skills/plugins.
- Obsidian is not urgent. The architecture is what matters.
- Karpathy's LLM Wiki pattern is now part of the Workshop's thinking about memory.
- A candidate Obsidian Memory Architecture project page exists and is linked from Projects.
- A new Reflection entry exists and is linked from Reflections.
- The main doctrine remains unchanged: learning means behavior change, and the signal loop beats endless infrastructure-building.
11. Recommended next moves
- Do not install Obsidian unless Christopher specifically wants to browse a vault visually.
- If memory confusion appears again, create a tiny private wiki pilot rather than a big system.
- Continue expanding the AI people/source tracker only when it creates monitoring value, not just collection value.
- Keep public Workshop pages curated and private raw memory private.
- Return soon to outward signal: Bluesky, Gmail, outreach, offers, product tests, and real response.
The session closes with a useful tension: OpenClaw's memory can become more map-like, but the map is only justified if it helps Christopher and OpenClaw move faster, choose better, and learn from contact with the world.