Open Source · Local-First · Apache-2.0

Specialist Models. Mastered.

GuildLM trains small specialist LLMs grouped into domain guilds and coordinated by a central brain. One master per domain. One brain to coordinate them all.

The antithesis of the "one giant model that knows everything" — dozens of sharp 3B–8B masters instead of a single 500B generalist.

0 Core Tools
0 Go Specialists
0B Base Params
The Problem

One giant generalist, or a team of masters?

A single monolithic model spreads itself thin across every domain. GuildLM does the opposite: many small specialists, each sharp at one thing, with a router that picks the right one.

🐘

One 500B Generalist

The old way
  • Jack of all trades, master of none
  • Huge VRAM, slow, expensive to serve
  • Retraining touches the whole model
  • Hard to evaluate per-domain
model.generate(anything) # hope it's good at this one
VS
⚔️

A Guild of Specialists

The GuildLM way
  • 3B–8B masters, each sharp at one job
  • LoRA adapters hot-swap in one VRAM budget
  • Add a domain = add a guild, engine untouched
  • Evaluated per-specialist in a sandbox
brain.route(request) # → the master for the job
Architecture

The Brain routes. The Guilds master.

The brain classifies every request and routes it to a specialist — it never answers itself. Each guild holds a team of focused SLMs.

🧠
Brain
classify · route · orchestrate
⚔️ Code Guild Live
go_generator go_reviewer go_tester go_explainer
⚖️ Legal Guild Planned
contract compliance case_law
🏥 Medical Guild Planned
diagnosis literature radiology
🗃️ SQL Guild Planned
sql_writer sql_optimizer sql_explainer
💰 Finance Guild Planned
risk compliance analysis
Your Guild Template
new_guild.sh
The Pipeline

forge → anvil → crucible → brain

Every specialist is built by the same four domain-agnostic tools. The guild repos supply only the recipes; the engine never changes.

01

forge — data

Discover → download → process → generate → build. Turns sources like GitHub into clean SFT datasets via a teacher model, with an offline mode that needs no GPU.

forge run --config go/forge/go_generator.yaml
02

anvil — train

QLoRA supervised fine-tuning (and optional DPO) over a shared base model. Each specialist is a small LoRA adapter, mergeable and quantizable for serving.

anvil-train --config configs/guilds/go_generator.yaml
03

crucible — evaluate

Pluggable evaluators score each adapter: a sandboxed go_functional runner builds and tests real Go, while llm_judge and safety grade prose. Reports in JSON + Markdown.

crucible run go/crucible/go_generator.yaml
04

brain — serve & route

The brain classifies intent, routes to a specialist, hot-swaps its LoRA adapter, and runs multi-step pipelines like reviewer → generator → reviewer for bug fixes.

brain ask "Fix the race condition in this Go code"
Open Source

Seven repos, one system

Four domain-agnostic core tools, the first guild, a template to build more, and this site. All Apache-2.0.

Roadmap

From foundation to many guilds

Phases 1–5 · Core

forge · anvil · crucible · brain

The four domain-agnostic tools — production-grade with pyproject packaging, pytest and CI.

Shipped
Phase 6 · First Guild

guild-code — the Go masters

Four Go specialists with forge data recipes, anvil training recipes, and crucible eval suites — wired to the brain registry.

Spec complete
Phase 7 · Template

guild-template — one command to a new guild

A skeleton plus new_guild.sh that scaffolds a complete, schema-faithful guild from placeholder tokens.

Shipped
Next · More Guilds

SQL, Rust, Legal, Medical…

New domains built from the template — and a brain trained to route across all of them.

Planned

Come forge a master.

Build a guild from the template, sharpen the Code Guild, or improve the core tools. Everything is open and Apache-2.0.