DesirePath
DesirePath is a local-first thinking partner that sits on top of a personal knowledge vault. It reads the notes you already keep, scores and connects them, and surfaces the ideas worth revisiting — without shipping your data to anyone else's servers.
How I use it
It is the workhorse that is the starting point of every project I create. It improves existing workflow by factoring in my past experiences, pain points, lessons, letting me create extremely detailed workflows in a single prompt.
When I want to create a new project, I'll just type in "Does this idea work?". The vault digs deep and acts like my thinking partner.
The problem
A knowledge vault grows faster than anyone can curate it. Most notes are never revisited, valuable connections stay invisible, and running a frontier model over every note on every change is prohibitively expensive. DesirePath treats curation as an infrastructure problem, not a manual chore.
How it works
Per-purpose model routing
Every task is routed to the cheapest model that can do it well. Embedding and tagging go to small local models; only genuinely ambiguous synthesis is escalated to a frontier model. Routing rules live in config, so swapping models is a one-line change.
Batch curation scoring
Rather than scoring notes reactively, DesirePath scores them in scheduled batches. Batching amortizes model warm-up, enables bulk discounts on hosted inference, and keeps the interactive path fast.
3-tier validation
- Tier 1 — heuristics. Cheap, deterministic checks filter the obvious cases.
- Tier 2 — small model. A local model handles the ambiguous middle.
- Tier 3 — frontier model. Only the residue that survives tiers 1–2 reaches an expensive model.
Dockerized deployment
The whole stack ships as a Docker compose file. Clone, docker compose up, point it
at your vault, and it runs entirely on your machine.
Results
Tiered validation keeps the vast majority of curation on cheap local compute, with frontier-model calls reserved for the handful of cases that actually need them.
Read the architecture deep dive Read the algorithm behind the name desirepath