Building 🧑💻 https://teapaw.org/t.co/dbvBoEpVap Organising 🤝 https://teapaw.org/t.co/85u71OJGdP
United Kingdom
Joined January 2022
- Tweets193
- Following195
- Followers73
- Likes153
Tried to explain what we actually mean by agents with memory. Most of them rewrite a file so they only know whats true today. ours keeps the old version with the date on it, which turns out to matter a lot for accountants
“we invest in pre product teams”
great, we have a product
“is anyone using it”
yes
“sorry, you havent fully solved GTM yet, we're out”
we're in LangChain's integrations. a few cards down from Microsoft, which I'll take 😅
langgraph-minnsdb is a LangGraph store that keeps every version, so your agent knows they lived in London before Berlin.
pip install langgraph-minnsdb
have a go and tell me if it breaks
Fun one today. Every deploy of our site broke every tab people already had open, because the deploy deleted the old build's files while browsers were still asking for them. Fix was catching the failed preload and reloading. Shipped.
MCPs are the death of Dashboards. First dashboards then UI in general. If I can connect an MCP to Cursor or Claude and ask “what changed last month?”, why do I need to know where the filters are?
Did it work? What failed? What changed? Tell me.
Dan Parry retweeted
your agent's memory isn't slow because of the algorithm. its slow because of how the graph is represented.
Adjacency lists + zero-allocation traversal = real-time agents that actually keep up 🏃
blog.minns.ai/zero-allocatio…
Dan Parry retweeted
We scored 70. The competition scored 27.
We benchmarked Minns on StructMemeval.
Minns outperformed other systems across all categories, with especially strong results in multi-hop state reasoning.
Even when paired with a stronger model, the competition only reached 62%
