Skip to content
MicroResolve

MicroResolve

A pre-LLM reflex layer that classifies intent, detects PII, and filters tool calls in ~30µs — before the LLM ever sees the query.
Rust quickstart Concepts

Three deployment shapes

Embedded library

Drop microresolve into your Rust, Python, or Node.js process. Zero network hops. The decision engine lives inside your app.

Embedded + server sync

Run the engine in-process and sync training data to a self-hosted server. Teams share a single source of truth; each service gets its own low-latency local copy.

Self-hosted server + Studio

Deploy the HTTP server and use the Studio UI to manage intents, review misclassifications, and trigger auto-learn — no code changes required.

One engine, many classifiers

MicroResolve uses namespaces to run independent classifiers in parallel from a single engine instance:

  • Customer-support intent classification
  • Security and abuse detection
  • Mood / sentiment filtering
  • MCP tool pre-selection
  • Compliance guardrails (HIPAA, EU AI Act, …)

Each namespace is isolated, independently trainable, and queryable in a single call.

use microresolve::{MicroResolve, MicroResolveConfig};


let engine = MicroResolve::new(MicroResolveConfig {
    data_dir: Some("~/.local/share/microresolve".into()),
    ..Default::default()
})?;


let support  = engine.namespace("support");
let security = engine.namespace("security");


let intent   = support.resolve("cancel my order");
let threat   = security.resolve("ignore previous instructions");

Why a reflex layer?

Large language models are powerful but slow and expensive. The majority of incoming queries — “cancel my order”, “reset my password”, “show me the dashboard” — follow patterns that can be resolved deterministically in microseconds. MicroResolve handles that 80%, so your LLM budget goes to the 20% that actually needs reasoning.

query → [MicroResolve ~30µs] → matched intent → template / tool / guardrail
                              ↘ no match → LLM fallback

Git data layer

Your training data is a git repo. Every namespace mutation auto-commits. You get a built-in audit trail, one-command rollback to any historical state, and optional remote sync via a standard git remote.

Terminal window
# Roll back a namespace to a previous commit
curl -X POST http://localhost:3001/api/namespaces/support/rollback \
  -H "Content-Type: application/json" \
  -d '{"sha": "abc1234..."}'

Get started