Local AI · Tool-first · Reproducible

Agents that
reason,
locally.

Logician is a local-first coding agent with a streaming terminal UI. SSH-ready, thinking-visible, and built for real code editing workflows.

View the docs Open source
🛠️ Tool-driven agents 📚 Repo grounding 📝 Streaming traces 🔌 MCP & skills 🧠 Persistent memory
01 — Capabilities

Why Logician?

Most agent frameworks hide what they're doing. Logician makes every tool call, trace, and reasoning step visible and inspectable.

🎯
Structured tool flow
Typed tools and explicit execution paths make agent behavior trivial to inspect, log, and debug — no black boxes.
🧠
Persistent context
Semantic memory and repo grounding keep answers tied to your actual work, session after session.
🔍
Repo grounding
Index repositories, build searchable code graphs, and let the agent reason directly over your real codebase.
🖥️
Local-first design
Runs entirely on your machine. Connect to any OpenAI-compatible backend or swap providers with a single config change.
📊
Research-ready
Built-in workflows for code review, literature synthesis, and reproducible investigation with audit trails.
🔌
Extensible skills
Plug new tools and domain-specific workflows on top of the stable agent core without rewriting anything.
02 — Quick start

Up in three commands

Clone, install, launch. The TUI connects to any OpenAI-compatible backend — configure once, start reasoning.

terminal — quick start
git clone https://github.com/lseman/logician.git
Clone the repo
cd tui && npm install
Install dependencies
npm start
Launch the TUI
03 — Skills

Extend with SKILL.md

Skills are plain Markdown with YAML frontmatter. Add domain expertise without touching the agent core.

SKILL.md 
---
name: code-review
description: Review code changes for correctness, risks, and missing tests.
aliases:
  - code review
  - review code
triggers:
  - review the code
  - review changes
  - check this PR
  - code review
preferred_tools:
  - read_file
  - grep
  - git
  - file_diff
example_queries:
  - review the latest changes
  - check this PR for bugs
when_not_to_use:
  - the user wants a style/formatting review only
next_skills:
  - cpp-code-review
  - python-code-review
  - javascript-code-review
---

# Code Review

Inspect changed files for bugs, regressions, and quality issues.

## Workflow

1. Identify changed files using `git diff` or `file_diff`
2. Read each changed file to understand context
3. Check for logic errors, missing error handling, and performance issues
4. Report findings prioritized by severity
04 — Design principles

Built for developers

Every design decision prioritises inspectability, local operation, and composability over convenience magic.

01
Tool-first architecture
Every agent capability is a typed, inspectable tool call — not a hidden chain.
02
Audit-ready logs
Every step is recorded. Replay, diff, and verify agent reasoning at any point.
03
Local or cloud
Start with llama.cpp, swap to any API backend with a single config change.
04
Composable skills
Extend the agent core without breaking existing workflows or tools.