The Infinite Memory Solution for Agents. Own your memories, use any harness, don't fear context collapse. No more .md file hell for agents, memories or skills.
Colorado
Joined March 2026
- Tweets488
- Following6
- Followers33
- Likes133
MentisDB retweeted
Your agents shouldn’t forget the decisions that make the next run better.
MentisDB is a durable semantic memory engine and versioned skill registry for AI agents that need persistent context across tools and sessions.
It helps you carry decisions, lessons, and instruction bundles forward by storing typed thoughts in an append-only, hash-chained memory log and keeping skill updates as immutable versions.
Key features:
• Shared memory across tools – connect MCP-capable coding environments to the same MentisDB daemon
• Context checkpoints – agents can write a Summary checkpoint, compact, and reload recent context
• Semantic memory – the optional local-embeddings feature enables semantic vector search with FastEmbed
• Versioned skills – uploads to an existing skill ID create immutable versions stored as unified diffs
• Verifiable provenance – agents with registered Ed25519 keys must sign their skill uploads
It’s open-source (MIT license).
Link in the reply 👇
MentisDB 0.10.10 is out 🧠
🐳 Official Docker image: docker run ghcr.io/cloudllm-ai/mentisdb
🐧 Linux binaries now run on Ubuntu 22.04 & Debian 12
🔒 Webhooks to your private network (Tailscale/WireGuard)
Durable, self-hosted memory for your AI agents.
github.com/CloudLLM-ai/menti…
PSA: private vulnerability reporting is now enabled in our Github repository.
Please send your notes through the repository's Security → Report a vulnerability form.
github.com/CloudLLM-ai/menti…
MentisDB solves this.
The memory graph can invalidate and point back at previously wrong assumptions or things that are no longer true, however, nothing is forgotten.
Memories are stored the same was as a blockchain, tamper proof.
AI memory sounds impressive
until it remembers one wrong thing
you said it once 6 weeks ago
the model treats it like law
and every new answer
builds around that old assumption
it still knows the files
the old name
the joke you made in passing
it does not know
you changed your mind
the hard part isn’t storing 10,000 things
it’s knowing which 10
should still matter tomorrow
MentisDB retweeted
Forked @mentisdb to play with letting my memory chains dream using a local ollama instance on a scheduled interval (to make it easier to wall off any unintended runaway token consumption). Got the basic hooks in place but I want to adjust temperature and generation params so it frequently has "working through a problem" dreams but occasionally it can have the equivalent of a "whoa is that from our meds?" dream.
Durable memory, invisible, your way.
Built for agents that learn.
Don't get locked in by any vendor, own your agent's memories.
Don't repeat yourself.
Let intelligence compound with experience.
Try it today, free and open
mentisdb.com
New MentisDB
This is a technical-debt release. It closes the two open bug reports filed against the 0.10.x line
- a vector sidecar bug that could make a chain stop opening (@FaulknerPearce Thanks)
- a CLI flag that was documented but not wired
New MentisDB 0.10.6 build 51 release
github.com/CloudLLM-ai/menti…
cargo install mentisdb --locked --force
Powered by Grok 4.6
