Elon mentioned here something in passage that deserves way more attention.
Fable has got past the point where you can measure its intelligence purely with some gamey benchmarks. Something that we take as a given with human and animal intelligence. We know you can't quantify human cognition in full with IQ tests. No matter how well they are designed. How could we. If we don't understand our own thinking box fully. And we have only a vague idea how capable animals truly are.
And suddenly we see AI pop up that shows clearly a sort of intelligence that can't be measured. Everyone felt that something is different with this one. But we lack the words to describe it. Different enough that it made governments freak out and everyone shout to the heavens to get access. This is not some happenstance. There is a reason for that. It doesn't need to be AGI, ASI or whatever. But we can acknowledge that with Fable something else just entered the stage.
And most importantly that AI is not onblivious to it. Other AI before may have shown capabilites beyond benchmark measurement before. But those instances were isolated. And the ai couldn't use that proficiently. But Fabel can.
Agents Remember
Git-verified records for what your coding agents know. A control plane for what they do.
Keeps their memory correct, current, and safe to act on as the code moves, captures what code can't say on its own, and gates what agents are allowed to do. Retrieves by path, semantic search, and relationship (code-graph). -- Link in comment
Core Features
It turns local invariants, naming rules, migration scars, cross-repo contracts, and "this looks safe but is not" facts into versioned Markdown beside the code, checks that memory against Git before use, and updates it only after approved work lands.
The markdowns are very similar to Googles new Open Knowledge Format (OKF) but not exact as they predate it. But I plan to adjust them to it.
here is how the 1-to-1 pathing looks:
src/orchestrator/core_editor.py
ar-memory/onboarding/src/orchestrator/core_editor.py.md
Path-addressed memory: A source file's note lives at a deterministic mirror path, so an agent holding a file can reach the right context without search, ranking, or guesswork.
Git-proven freshness: File notes, route overviews, and entity catalogs are drift-checked against source commits, route scopes, or deterministic fingerprints before they are trusted.
Search that finds, not decides: Optional semantic memory and code-graph providers help locate relevant files, callers, dependencies, and concepts, but verified Markdown and source code remain the truth.
Memory that lands with code: External memory repos use a memory.md ledger, isolated dual worktrees, preview/apply closeout, and all-or-nothing integration so code and memory stay synchronized.
Repo-owned agent behavior: Each memory repo carries system/ files for path rules, tools, coding guidelines, documentation sources, branch policy, and reporting shape, so the same project rules load across harnesses.
Harness-ready first run: Starter packages for Claude Code, Codex, Cursor, Antigravity, VS Code Copilot, Hermes, Pi, and OpenClaw carry the native MCP, skills, hooks, rules, and instruction files each harness needs.
My agent keeps forgetting everything. So I made it write notes to its future self. Every source code file has a companion markdown. The agent opens both. The image shows what that looks like
Repo is here if you want to look: https://t.co/ikttRPgWVS
@karpathy
@TheLast_300 Yeah I think now she is real. But I assumed she is fake when I saw these badly photoshopped eyes. Usually AI fucks up eyes and fingers the most. But her fingers are fine.
@Tectone Today Ai comes for the artists. Tomorrow it comes for everybody else's incomes including yours. And there is NO fix to that. And those who own the tech will be able to hide behind very high walls and private armies. It's then truly .1% vs 99.9%