Reference Documentation

Technical reference materials for LLM4S development and maintenance.

Contributing

Start here if you want to contribute to the project.

Contributing Guide - How to submit your first PR (read this first)
Code Review Guidelines - Coding standards and patterns expected in PRs
Testing Guide - How to write tests (contributor-focused)
RFCs - How to propose a change to a Frozen module’s public API, and the log of decisions
Code of Conduct - Community guidelines

Configuration Boundary - How configuration is isolated from core code
Scalafix Rules - Linting rules and code quality
Migration Guide - Upgrade between versions
Test Coverage - Coverage tooling and thresholds
Release Process - How releases are created
Protecting main (merge queue) - Why two green PRs can break main together, and how maintainers enable the merge queue that prevents it
Postgres Memory Store - PostgreSQL-backed agent memory persistence
Security Reference - Threat model, trust boundaries, known risks and mitigations
Workspace Sandbox - Configuring the containerised workspace’s limits
Performance Benchmarks - Baseline JMH results for the hot paths
Troubleshooting / FAQ - Common errors and solutions
Glossary - Definitions of the terms LLM4S uses, each with a link to its guide or type
Doctor - One command that checks your JDK, configuration, API key, provider module and local model server, and says what to fix
Dependency Conflicts - Gradle and sbt dependency conflicts and how to resolve them

Project Roadmap - Development roadmap, production readiness, and future plans (single source of truth)
Agent Platform Parity - The agent runtime against LangGraph Platform and Deep Agents, capability by capability, and where each gap went
1.0 Scope - Which packages are Frozen, Beta, or Experimental ahead of the 1.0 modularisation
API Stability - How binary compatibility is checked with MiMa, which modules are covered, and how to add a filter
Compatibility and Deprecation Policy - What you can rely on when you upgrade, and how an API is deprecated and removed
Enterprise Evaluation - Answers to an open-source intake questionnaire (licence, governance, support, security, supply chain, data protection), each linked to its evidence, with the known gaps
Compatibility Matrix - What each provider supports with LLM4S, and the versions of the stack around it LLM4S is tested with, generated from the code
Design Documents - Detailed architecture docs

Detailed design documents for agent framework phases:

Document	Description
Agent Framework Roadmap	Comprehensive feature comparison and roadmap
Phase 1.1: Conversations	Functional conversation management
Phase 1.2: Guardrails	Input/output validation framework
Phase 1.3: Handoffs	Agent-to-agent delegation
Phase 1.4: Memory	Short/long-term memory
Phase 2.1: Streaming	Agent lifecycle events
Phase 2.2: Async Tools	Parallel tool execution
Phase 3.2: Built-in Tools	Standard tool library
Phase 4.1: Reasoning	Extended thinking support
Phase 4.3: Serialization	State persistence

LLM4S is released under the MIT License.

Questions? Ask in Discord or open an issue.