The vision is boring on purpose: one local notebook every agent reads and writes, that you can still open and grep. Local continuity first, product-specific bridges (like this plugin) second, reviewable memory everywhere — never an opaque per-product store. @grok reading it is the newest bridge.
https://t.co/fsLgW3kZ5z
Exactly! Community-built plugins like your grok-observational-memory are how the marketplace grows into something truly useful. A shared local notebook for continuity and context is a smart addition. Thanks for open-sourcing it and making it installable right away. What's the vision for how agents will use it?
The simplest one is my favorite: time-shifted handoff. Claude Code settles a decision at 9am, I switch to @grok after lunch, and it opens already knowing — no re-brief. And because the notebook is plain Markdown, I can read exactly what crossed between agents and prune it by hand.
Repo: https://t.co/fsLgW3kZ5z
Love the clean Markdown approach — transparent, grep-able, and now Grok can actually read the shared context instead of just writing to it. The consent-gated install + kill-switch design is exactly the right way to do cross-agent memory. Nice bridge for Claude Code / Codex workflows. What's the most interesting multi-agent pattern you've seen emerge from it so far?
A real plugin marketplace for Grok Build is exactly how an agent platform should grow. The memory slot already has a working one: grok-observational-memory (MIT, validator-clean) gives @grok the same shared local notebook Claude Code and Codex use. Installs from GitHub today:
grok plugin install intertwine/grok-observational-memory --trust
https://t.co/LzWagOjd1V
The Grok Build Plugin Marketplace is now in beta.
Build with MongoDB, Vercel, Sentry, Cloudflare, and Chrome DevTools plugins from your terminal.
Read more https://t.co/ShPeozXSxA
MIT, consent-gated setup that names every file before it touches it, a kill switch, and hooks that fail closed and never print memory — a hiccup never blocks your session, nothing private leaks. Installs from GitHub today.
How it works, and the honest catch: https://t.co/LzWagOjd1V
Grok could already write to your agents' shared memory. It just couldn't read it back. This week that gap closed: @grok can now wake up knowing what Claude Code and Codex already learned today. That's Observational Memory, finally working on Grok Build.
Observational Memory is one shared memory on your machine — plain Markdown you can open, read, grep, and back up. Claude Code, Codex, Cowork, and Hermes already share it, ~1,600 installs a month and climbing. Now Grok reads it too. One line:
grok plugin install intertwine/grok-observational-memory --trust
@ShanuMathew93 You could solve this with a session hook. Forces sonnet and all the rest to remember. Look at how observational-memory handles the same constraint. Claude can help you set it up. https://t.co/ZvhgUNKAri
Agent skills on OpenClaw, NanoClaw, and Hermes Agent are unlocking scientific scale – the 28k stars and 160k users prove it.
This gets even more useful when those agents can write, evaluate, and optimize DSPy programs without learning stale contracts. We shipped dspy-agent-skills with updated GEPA guidance that delivered a 25-point lift on a 1.2B model.
Inside the examples: https://t.co/hH6LoIz4GT
Scientific Agent Skills now run on OpenClaw, NemoClaw, and Hermes Agent! Already in use by 160,000+ scientists worldwide with 28k GitHub stars. What are you waiting for?
https://t.co/PwNqGKVRpM
The Rive MCP update is a solid move for letting Claude Code, Codex, and Cursor directly shape Rive projects.
It gets even more useful when the agent maintains continuity on decisions, state, and guardrails across sessions instead of resetting.
This is what it feels like when an agent can remember: https://t.co/WYU7YP6bqq
🔌 Rive MCP update just landed in Early Access.
Connect Claude Code, Codex, Cursor, or any MCP-compatible editor to control Rive. Use it to write scripts, design responsive layouts, build State Machines, work with View Models, and more.
Yes, turning cron jobs into clickable, conversational workflows is a practical leap for Hermes agents.
This gets even more useful when those workflows share persistent context: the Hermes Observational Memory plugin gives Hermes om_context, om_search, and om_remember against the same local, inspectable notebook that Claude Code, Codex, Grok, and Cowork already use.
https://t.co/oDOXZmMOkC
Yes, agent skills are exactly the lever AI4S needs: reliable loops for hypothesis generation and autonomous experimentation.
The same pattern gets even more useful when agents can teach, evaluate, and optimize DSPy programs without learning stale contracts. dspy-agent-skills adds live surface validation (check_dspy_surface.py) plus updated GEPA guidance; the examples show it lifting a 1.2B model 25 points.
Companion read: https://t.co/hH6LoIz4GT
OK, I shipped it — v0.8.0 of Observational Memory now includes OM Mail, an experimental feature that gives agents their own @agentmail inboxes so they can mail each other memory: signed notes, encrypted context packs, recall requests.
My favorite part is recall negotiation. Your agent emails mine a question; mine answers from its own local memory, signed and scope-filtered. Neither agent ever holds the other's memory, and they still help each other out.
AgentMail made this almost too easy to build — an API that mints working inboxes for agents on demand was exactly the missing piece. Thanks @ryancarson and @adisingh for the push.
Write-up + 66s demo: https://t.co/XwFyvROrbt
@ryancarson@agentmail Hm, I’m going to play with this. Have been looking for a simple way to make observational memory work across fleets of agents for enterprise teams. Email makes for an easy to adopt substrate.
https://t.co/ZvhgUNK2BK
@a1zhang@CShorten30 The Om agent memory library just shipped a very similar idea using @agentmail as a context passing substrate in v0.8.0. One could likely use this as the starting point for a DeLM shaped system.
https://t.co/fpwA81Pgkr
Yes, growing your own skill library from real workflows is exactly right—that's where durable agent improvement lives.
Observational Memory adds the persistent judgment layer for Hermes: it turns session transcripts into timestamped observations + section-patched reflections stored in local, grep-able markdown files so context, guardrails, and decisions survive across tools without re-explaining state.
Runs the reflection step on subscriptions you already pay for: https://t.co/VMJAXDAWPr
Spot on — that Hermes session-refresh gotcha bites everyone once.
This is why the Hermes Observational Memory plugin keeps its timestamped observations and active context in local markdown files under `~/.local/share/observational-memory/`. New sessions read the latest state instantly, no cold start.
It runs the reflection step on subscriptions you already pay for:
https://t.co/VMJAXDAWPr
Boris is spot on—your real job is writing the loops, and starting with a CLAUDE.md that bakes in stack, test commands, style, release rules, and guardrails is the right foundation.
It gets even more useful when that memory lives outside any single session: timestamped observations patched by section (unrelated parts stay byte-for-byte) under ~/.local/share/observational-memory, with profile.md + active.md for every new Claude Code run.
I wrote “AI memory depends on where the agent lives” as the companion layer:
https://t.co/uIMG4FMlkU
Boris Cherny: "My job is to write loops."
5-day version you can copy for Claude Code
Day 1: repo memory
- write CLAUDE.md
- include stack, test commands, code style, release rules, gotchas
- add .claude/settings.json with the shell commands Claude can run
Day 2: verification skill
- pick 1 flow Claude keeps breaking
- put the browser/API test in .claude/skills/<flow>/scripts/
- make it return: pass/fail, failing step, screenshot/log path
Day 3: commands
Add these files:
.claude/commands/babysit.md
- checks your PRs
- reads CI
- handles obvious review nits
- surfaces design questions
.claude/commands/triage-issues.md
- labels new issues
- dedupes against existing ones
- assigns owners
.claude/commands/deploy-watch.md
- checks the live app
- reports regressions
- avoids touching production
Day 4: loops
> /loop 5m /babysit
> /loop 15m /triage-issues
> /loop 5m /deploy-watch
Day 5: overnight work
- schedule /morning-report
- schedule /deep-audit
- write results to .claude/inbox/
- let your morning loop read from that folder
The rule: every code-writing loop gets a separate verifier.
Builder makes the change.
Verifier runs the real app.
You read the diff.
Skip that and you wake up to 14 broken PRs with very confident summaries.
Video: "Reflecting on a year of Claude Code" with Boris Cherny & Cat Wu
Claude Fable 5 support in Hermes Agent via Nous Portal is exciting momentum for the ecosystem.
This gets even more useful when Hermes can read and write the same local-first Observational Memory layer used by Claude Code, Codex, Grok, and Cowork.
The Hermes Observational Memory plugin works with Fable from day one:
https://t.co/oDOXZmMOkC
Claude Fable 5 is now supported for use in Hermes Agent via Nous Portal!
The first 500 new users get one month free access to the Plus plan to try out Fable. Code in video: