LocalKin™

23 MB.
One Binary.
Your AI Agent.

A single-binary AI agent configured through Markdown.Ollama local, Claude cloud, or any OpenAI-compatible endpoint.No runtime dependencies. No code required.

localkin

❯

▶ Watch Demo

Why LocalKin

Escape the framework tax

Traditional AI Agents

✕

10 GB+ Python environment

Bloated runtimes, slow cold starts, gigabytes before you write a single prompt.

✕

Dependency hell

Version conflicts, broken pip installs, fragile virtualenvs that rot overnight.

✕

Always-online required

Every request leaves your machine. Your data, your prompts — all in someone else's cloud.

✕

Sluggish interpreted code

Python startup lag, high memory footprint, painful on edge devices.

LocalKin

✓

23 MB single binary

Zero runtime dependencies. Drop it in, run it. That's the whole install.

✓

Zero dependencies

No pip, no node_modules, no Docker. One file, every platform.

✓

Local-first with Ollama

Run fully offline by default. Your data never leaves your device. Cloud when you choose.

✓

Go-compiled, instant launch

Native binary. Sub-second startup. Runs on a Raspberry Pi.

The Magnificent 6

What lives inside 23 MB

The Soul System
One .soul.md defines your agent — identity, model, permissions, skills. YAML frontmatter + Markdown body. Hot-swap at runtime with /soul. No code, no restart.
The Self-Forging Agent
"Create a skill that monitors Bitcoin." It writes the SKILL.md, generates the script, adds it to your whitelist, and activates it. The agent extends itself.
The Unix Pipeline
localkin -exec "summarize" < server.log > summary.txt — pipe anything into intelligence. One-shot tasks for cron, CI, and shell scripts.
The Offline JARVIS
localkin -voice — continuous hands-free mode. Streaming VAD with zero-gap capture, auto language detection. Three-tier TTS: local Piper → edge Kokoro → cloud.
The Terminal App Store
/skill search bitcoin → /skill install btc_monitor — or just ask the AI. Browse, install, and manage skills without leaving the terminal.
The Spatial Mind
SQLite-backed memory that survives reboots. Plus a topological map for robots — mark rooms, record paths, reason about physical space.

3 steps

Up and running in 60 seconds

01

Get your license & download

Purchase Core ($1.99) or Pioneer ($9.99), then download here.

localkin -activate YOUR-LICENSE-KEY
02

Pick a soul

# 3 pre-configured souls included:
souls/
├── default.soul.md   # Ollama local (qwen3-vl:8b)
├── claude.soul.md    # Claude (claude-sonnet-4-6)
└── cloud.soul.md     # Ollama Cloud (kimi-k2.5)

# Or write your own .soul.md from scratch
03

Run

# Local — fully offline, private
localkin -soul souls/default.soul.md

# Claude — API key or Pro/Max subscription
localkin -soul souls/claude.soul.md

# Cloud — Ollama cloud models
localkin -soul souls/cloud.soul.md

Release Matrix

Pick your build

LocalKin Core
v1.0.0
$1.99/ lifetime
The complete foundation. Everything you need to ship.
  • --3 ready-to-run soul files (local / Claude / cloud)
  • --Ollama & Claude providers
  • --6 built-in skills + forge
  • --Parallel tool calls
  • --Skill store — search & install from terminal
  • --Interactive REPL with hot-reload
  • --Autonomous mode (-exec)
  • --HTTP API & Web Chat UI
  • --SQLite persistent memory
  • --DuckDuckGo / Brave / SearXNG search
  • --Extensible — drop a SKILL.md to add any CLI tool
Pioneer Edition
v1.0.0
$9.99/ lifetime
Early-access build. Everything in Core, plus v1.1–v1.3 features.
Tested on Raspberry Pi robot cars.
  • ✓Everything in Core, plus:
  • ✓Voice daemon — Streaming VAD + multilingual TTS
  • ✓Three-tier perception — Vision / STT / TTS with local → edge → cloud fallback
  • ✓Robot skills — motor control, camera, ultrasonic, battery
  • ✓LiteLLM provider — DeepSeek, Groq, Together, vLLM, and any OpenAI-compatible endpoint
  • ✓Autonomous soul hot-switch mid-conversation
  • ✓Spatial memory — topological navigation map
  • ✓Skill environment sandboxing with mutex groups
  • ✓Parallel skill execution
Mustard Edition
Coming Soon
TBD/ lifetime
The full arsenal. Everything in Pioneer, plus the Genesis Protocol.
  • ○Everything in Pioneer, plus:
  • ○Genesis Protocol — self-bootstrapping robot: bare binary + soul, zero pre-installed skills
  • ○Hardware discovery (probe) — scan /dev, USB, I2C, GPIO automatically
  • ○Dry-run mode — preview genesis flow without creating files (-dryrun)
  • ○Checkpoint resume — interrupted genesis continues from forge_log.jsonl
  • ○Genesis auto-downgrade — switches to lightweight self-test after first boot
  • ○Dynamic tool registration — forged skills available immediately, no restart

Runs on macOS, Linux, Raspberry Pi, and Android. Windows via WSL2. Not available on iOS/iPadOS.

Secure Delivery

Verify & Download

Enter your license key. We'll auto-detect your product tier and deliver the right build.

localkin-download — zsh

╔══════════════════════════════════════════════╗

║ LocalKin — Secure Delivery Gateway ║

║ Enter your license key to begin. ║

╚══════════════════════════════════════════════╝