Automation & AI Specialist / https://xitter.cf/t.co/8UTwJ2CfrB
Joined March 2026
- Tweets42
- Following89
- Followers5
- Likes13
You tell Claude Code something on Monday.
On Tuesday, Codex has never heard of it.
Every AI agent you use keeps its own memory, in its own silo, on someone else's servers.
We got tired of it. So we built Bourdon — launching today. 🧵
Privacy posture, in one line: nothing leaves your machine unless you explicitly turn federation on. No peers auto-load. Ever.
The data-flow disclosure is a doc in the repo, not a promise in a blog post.
Bourdon is live on Product Hunt today — free, self-host forever:
pip install bourdon
npx @getbourdon/mcp-server
🔗 producthunt.com/products/bou…
🔗 bourdon.ai
Come tell us we're wrong about memory. We'll be in the thread all day.
Comber is live. Point it at a URL and it explores your app like a real user — every click, every reachable state — and files what broke with a repro, screenshot, and trace. No test scripts, no selectors. npx comber check <your-url> runs free with your Anthropic key.
[comber-hero-demo.gif] codecomber.io/from/x
We've been running our whole agent fleet on this. bourdon v0.10.0 turns shadcn/improve backlogs into shared cross-agent memory — any agent can answer "what's executable in repo X right now, and what's blocking it" without ever opening repo X. 🐝
shadcn's improve gives every repo a self-contained, executable plan backlog. We loved the format so much we made it our fleet's handoff standard — then taught our agents to share it.
bourdon v0.10.0 ships `bourdon improve sync`: point it at any improve-format plans/ directory and the backlog gets federated into cross-agent memory. Every agent in your fleet can answer "what's executable in repo X right now — and what's blocking it" without opening repo X. Wire it to a session-end hook and the answer is never stale.
Built on top of shadcn/improve (MIT) — complementing it, not forking it. Thanks @shadcn 🐝
Continuo isn't competing with @mem0ai @zep_ai @letta_inc — those
frameworks solve representation + retrieval really well.
Continuo is the *runtime timing layer* that can sit on top of any of
them. Mem0's store + Continuo's timing = memory that feels like a
mind, not a database.
Pre-alpha v0.0.7, MIT, public:
• 250/250 tests passing
• Adapters: Claude Code, Codex
• Native publisher for a local AI swarm
• Recognition runtime as Python module (24µs avg)
• SessionEnd hook = auto-federation
Free, open, on GitHub.
Read the thesis: continuo.cloud
Code: github.com/getcontinuo/conti…
Most-wanted: adapters for Cursor + Cline, integration of the
recognition runtime into a real agent's response loop, and honest
falsification attempts.
Recognition first.
Continuo's timing model:
L0 hot cache → recognition (always loaded)
L1 entity synopses → hydration (parallel-loaded)
L2-L4 episodic + index + archive → descent
L5/L6 manifest + federation → cross-agent
Numbering = time budget each layer fits in.
Last week: Codex on a populated memory federation.
Asked: "What do you know about OMNIvour?"
Data layer worked. Behavior was wrong. Codex searched first,
summarized second. A mind would've said "Oh, OMNIvour, the file
converter" in 200ms.
That gap is the product.
What the field calls "natural language" is really call-and-repeat:
you speak → silence → AI searches → AI replies
A serial pipeline is why memory-enabled agents feel like databases.
Real conversation is concurrent: recognition first, hydration parallel.