MicroResolve

MicroResolve is the System 1 relay for LLM apps. Every request runs through a sub-millisecond reflex layer that picks a candidate intent + confidence band and hands the result to your System 2 — your LLM, or a human reviewer for high-stakes domains (HIPAA, legal, financial). We never talk to your users; we give your decision-maker a head start.

Tool selection, intent triage, guardrail dispatch, refusal classification — the routing decisions your LLM keeps making run in 50µs here and improve on your traffic via corrections.

In the box

• Studio — web UI for namespace management, simulation, review, training. Git-backed history & rollback.
• Library — Python / Node / Rust. Embed in prod, or stay live-connected to a Studio.
• Online learning — Hebbian + LLM-judged corrections. No fine-tuning, no restart.
• Native imports — MCP, OpenAI functions, LangChain tools, OpenAPI specs.
• Multilingual — Latin + CJK tokenization; learns whichever language your traffic is in.
• One engine, many namespaces — tool routing, intent triage, compliance gates, abuse detection — same engine, isolated namespaces.

Not a perfect one-stop classifier. Every router has false-positive / false-negative failure modes; MicroResolve does too. The pitch is: fast deterministic routing that converges on your traffic via corrections — the corrections are the product. Cold-start packs give you a starting point; deployment-time corrections shape the index to your domain. Used right, it works.

v0.2 — early release; pin exact versions in production.

Documentation · Benchmarks & methodology · Changelog · Contributing

"the dispute on order 1834 came back invalid — refund the customer,
 mark the dispute resolved, then post in #ops that we lost it"

  → create_refund        (confirmed, high)
  → update_dispute       (confirmed, high)
  → slack_send_message   (confirmed, medium)
  → relation: sequential

                          ~92 µs · narrowed 129 tools to top-3
                          for the LLM, with no LLM call yet

LLMs do everything: reason, compose, generate, choose tools, handle context. MicroResolve only handles the repeated routing decisions — which tool, which intent, what's PII, is this safe. It learns continuously from your existing LLM calls — every misroute the LLM corrects flows back into the index, so the candidate set sharpens from production traffic alone.

Use it for:

• LLM-agent tool prefiltering — narrow 100+ tools to the top 3 the LLM actually needs to see this turn. Cuts prompt tokens, cuts latency, scales to large catalogs.
• Customer-support triage — route incoming tickets / chat messages to the right queue or workflow before the LLM gets involved (or without an LLM at all for the easy cases).
• Intent classification — single-utterance bucketing for conversational interfaces, IVR-style menus, search-query routing.
• Slash / chat command routing — recognize the user's command from free-form phrasing without retraining a model every time the catalog changes.
• Workflow / decision routing — multi-intent decomposition with relation detection (sequential / conditional / parallel / negation) for steps that need to fan out or chain.
• Permission and risk gating — classify a request into a risk tier before paying for an LLM round-trip.

Install

Python

pip install microresolve

Node.js

npm install microresolve

Rust

cargo add microresolve

Studio (single-binary UI + HTTP server)

Pre-built tarballs for Linux (x86_64 / aarch64, glibc + musl), macOS (x86_64 / aarch64), and Windows ship on every release.

# Linux x86_64 — adjust for your platform from the releases page
curl -L https://github.com/gladius/microresolve/releases/latest/download/microresolve-studio-x86_64-unknown-linux-gnu.tar.gz \
  | tar xz
./microresolve-studio --data ./data
# Studio at http://localhost:4000

All artifacts come from the same source-of-truth Rust core — same algorithm, same data files, fully interchangeable.

30-second demo

from microresolve import MicroResolve

engine = MicroResolve()
ns = engine.namespace("agent")
ns.add_intent("billing:cancel_subscription", ["cancel my subscription", "stop my plan"])
ns.add_intent("billing:request_refund",      ["refund my order", "i want my money back"])

matches = ns.resolve("end my subscription right now")
print(matches[0].id, matches[0].score)
# billing:cancel_subscription 0.96

Same shape in Node (engine.namespace(...).resolve(...)) and Rust (ns.resolve(...)). For the full pipeline including LLM-judged auto-learn, multi-intent decomposition, and Studio inspection, run the Studio binary and open http://localhost:4000.

Why MicroResolve

[!IMPORTANT] MicroResolve is a prefilter, not an absolute classifier. It returns ranked candidates — the LLM is the final picker. Your agent prompt must include a confirm_full_catalog fallback tool so the LLM can reach the full catalog when none of the candidates fit. That fallback is part of the design, not an optional safety net. See Confirm-turn pattern.

Benchmarks

Agent tool routing — 129 real tools imported from 5 production MCP servers (Stripe / Linear / Notion / Slack / Shopify), single namespace, scored as a prefilter for an LLM picker:

+ auto-learn = incremental Hebbian + LLM-judged phrase ingestion from production corrections. No retraining pipeline; the data updates in place. Reproduce: python3 benchmarks/agent_tools_bench.py (~$0.55 / ~10 min on Haiku 4.5).

Methodology, datasets, and reproduction scripts live in benchmarks/. What's not yet benchmarked: out-of-scope rejection (see Confirm-turn pattern), adversarial robustness, drift over multi-week production traffic.

Single-utterance intent classification (academic baselines):

How it works

Queries go directly to the L2 intent scorer — a deterministic IDF-weighted inverted index with online Hebbian learning:

raw query
  → L2  intent scoring  (IDF-weighted sparse term graph + Hebbian learning)
  → L4  cross-provider tiebreak  (when equal-score providers compete)
  → result

failures → review queue → auto-learn worker → L2 updated

L2 is the core scorer: each intent has a learned sparse vector of IDF-weighted terms. Multi-intent decomposition uses a token-consumption pass — once the top intent is confirmed, its tokens are consumed and the remaining tokens are re-scored to detect additional intents in the same query. L4 breaks ties when the same action exists across multiple providers. There are no preprocessing layers; raw tokens are looked up directly, making the engine safe for medical, legal, and code namespaces.

Every confirmed routing reinforces term weights. Every correction shifts weights away from the wrong intent. No retraining; the data updates in place.

Confirm-turn pattern (System 1 → System 2)

MicroResolve returns ranked candidates; the LLM picks one or falls back to the full catalog when nothing fits. Always include a confirm_full_catalog fallback tool in your agent prompt — without it, out-of-scope queries and novel phrasing route to the wrong tool:

Candidate tools (from the prefilter):
  - tool_a  (...description...)
  - tool_b  (...description...)
  - tool_c  (...description...)

If one of these clearly applies, call it. Otherwise — unsure, out of
scope, or candidates look unrelated — call `confirm_full_catalog` to
receive every tool, then pick.

The prefilter shrinks 150 tools to 3 in ~60 µs; the LLM is the final picker. Every confirmed call flows back as a corrected example, so the candidate set sharpens over time.

HTTP API

Send X-Namespace-ID: my-namespace to isolate intents per namespace. The core endpoints are /api/resolve, /api/intents, /api/training/{review,apply}, and /api/import/mcp/{search,fetch,apply}. Full reference in the server API docs.

Multi-intent, projection, and multilingual

Native multi-intent — a single query can confirm several intents in one call, with the relation between them detected:

"cancel my order and update my address"
  → cancel_order   (confirmed, high)
  → update_address (confirmed, medium)
  → relation: parallel

Detected relations: sequential, conditional, negation, parallel.

Projected context — co-occurrence tracking discovers what auxiliary intents typically fire alongside the primary one, so your orchestrator can pre-fetch them in parallel without an extra LLM round-trip:

"I want a refund"
  → refund_order  (confirmed, high)
  → projected:    check_balance (21 %), warranty_lookup (13 %)

These relationships emerge from accumulated routing — they're not configured.

Multilingual — Latin scripts via whitespace tokenization, CJK (Chinese / Japanese / Korean) via Aho-Corasick automaton + character bigrams, all in the same namespace. Per-language seed phrases per intent. Multi-intent decomposition runs after tokenization, so a Chinese query like "取消订阅然后退款" decomposes the same way an English query does.

Import formats

• MCP tools/list (and Smithery registry passthrough)
• OpenAPI / Swagger — each endpoint becomes an intent
• OpenAI function-calling schemas
• LangChain tools

Reference packs

Four pre-curated packs ship in the packs/ directory. Drop into any data dir and the engine picks them up on next start (or copy at runtime via engine.namespace(id) after staging the files).

All packs are bag-of-words IDF inverted indexes with a per-namespace voting-token gate (min) and threshold (thr). Pack format = directory of JSON files, one per intent + _ns.json for namespace metadata. Install by copying into your data_dir. They improve on your specific traffic via the correction loop — the cold-start numbers above are the floor, not the ceiling.

Each pack ships with a calibrated default_threshold and default_min_voting_tokens in _ns.json. That's the starting point for your traffic. Tune live in the Studio sidebar (TuningPanel) or via PATCH /api/namespaces for your FP/recall trade-off.

Position in the stack: MicroResolve is the System 1 relay — sub-millisecond, deterministic, running on every request. The result is always handed to your System 2 (typically your LLM; a human reviewer in regulated domains). System 2 acts on confident matches cheaply, or asks the user to confirm with the candidate list when we flag uncertainty. The published FP rates are sized for that architecture, not for unsupervised use.

License

Dual-licensed under MIT or Apache-2.0 at your option — the standard Rust ecosystem licensing. Both are fully permissive and allow commercial use.

• LICENSE-MIT
• LICENSE-APACHE — adds an explicit patent grant

Contribution

Unless you state otherwise, any contribution intentionally submitted for inclusion in this work shall be dual-licensed as above, without any additional terms or conditions.

Commercial support

Maintained by Gladius Thayalarajan — consulting on custom integrations, multilingual tuning, and agent-stack tool routing: gladius.thayalarajan@gmail.com.