Repository navigation
How I run mcp-memory-service across three machines and a VPS #1427
filhocf
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
I run a small fleet of AI agents — one coding/ops agent that follows me across three machines (two at home, one at work), plus two on a VPS. The service is the shared memory that makes "one agent on three machines" actually behave like one: I start a task on the laptop, continue it on the desktop, finish it at the office, and the agent carries the decisions, the mistakes, and the context forward because the memory syncs underneath. Crashes and context resets stop being events — the checkpoint is already in memory.
Day to day I lean on three things: semantic recall in Portuguese (I run the multilingual MiniLM model via ONNX, no torch at inference), session harvest that turns my work sessions into durable memories automatically, and quality/rating so the signal survives while noise decays. The live database is ~23k memories and growing.
The real change is in throughput. With memory carrying the context, I run several sessions in parallel — different problems, different repos, different machines — and the agents keep each one coherent without me re-explaining anything. Honestly, the bottleneck is no longer the tooling; it's my own cognitive limit in supervising that many parallel threads. That is a good problem to have, and it only exists because the memory layer holds.
Everything I've contributed back started there: I use it hard enough to keep hitting the edges, and each edge became a PR. The direction I want to take it next — self-maintaining memory (autolearn/autodream), server-side lifecycle, per-agent pluggable ingestion — is written up as RFCs for exactly this reason: the daily use keeps showing me what's missing. Two of those threads: pluggable, per-agent ingestion (#1393) and the background enrichment trilogy (#1286).
All reactions