Skip to content

⚡ UnicoLab · open harness · summer energy

Coding agents that still work when tokens run out ☀️

A premium coding harness tuned for SLMs — with the ambition of frontier tools — and a clean plug for any OpenAI-compatible LLM. Plan, specialize, criticize, learn. Less “paste the monorepo”. More “ship the tiny win”.

🚀 Install now ⏱️ 60-second start 🧠 How it thinks

🍎 oMLX default 🖥️ Premium TUI 🎨 Offline Studio 🔌 Any provider 💚 UnicoLab

🦞

Frontier agents are wonderful — until the meter spins like a washing machine and your context window becomes a landfill of half-read files. SLMCode is the harness that says: “maybe don’t feed the whole turkey to a 7B model.”

Why this exists 🤔

Frontier coding agents are wonderful — until the invoice arrives wearing a tuxedo.

Small local models need a different loop: atomic tasks, evidence gates, markdown memory, and a critic that looks at the disk, not just the vibes.

🎤 The pitch in one breath

Keep the ambition of Claude Code / Antigravity / Pi-style harnesses. Fill the gaps so 7B–30B models (and bigger ones) can ship — maybe with longer passes — without pretending they can swallow your monorepo whole. 🦃

😵 Dump the repo

  • Giant prompts
  • Sleepy SLMs
  • Hallucinated “fixes”
  • Mystery diffs

😎 Scoped packs

  • Tiny TaskPacks
  • Specialists with jobs
  • Disk-backed critic
  • Memory that compounds

Start here 🗺️

  • 📦 Install

    One-liners for macOS, Linux, Windows + Homebrew. No Go required. Your PATH is about to feel special.

    Get on PATH

  • 🚀 Quick start

    Init a playground, run a tiny task, open Studio. Feel the loop. First green run ≈ one coffee sip.

    First green run

  • 🧠 Concepts

    Why scoped packs beat “paste the repo”. Explore reuse. Self-critic. Flywheel.

    Mental model

  • 🔌 Providers

    oMLX, Ollama, OpenAI, OpenRouter, vLLM, or your private gateway. Bring your own brain. We bring the loop.

    Plug a brain in

  • 📘 Handbook

    Daily guide, TUI, skills, Studio, agents, battle-tested recipes. Less theory. More “paste this and win”.

    User guide

  • 📚 Reference

    Full CLI, config.yaml, FAQ, and “why is doctor red?” triage. For when things get spicy. 🌶️

    CLI reference

The loop (30 seconds) 🔁

flowchart LR
  Q[Query 🎯] --> P[Plan + split 📋]
  P --> C[Coordinator 🧭]
  C --> W[Workers 🛠️]
  W --> R[Reviewer 🔍]
  R -->|reject 😬| X[Corrector 🔧]
  X --> R
  R -->|approve ✅| L[Learn + skills 🦋]
  L --> M[(.slmcode memory 💾)]

Routing stays in Go. Models get tiny packs. Memory compounds on disk. You stay in the TUI or Studio — not in a black box labeled “trust me bro”.

What “premium” means here ✨

Promise Reality check
👀 Live visibility Agent, scope, patches, latency — not a spinner cult
🛟 Recoverable runs /stop → checkpoint → /resume
🛡️ Safety rails dry-run / review / shell allow·ask·deny
🔌 Model-agnostic Same harness for local SLM or cloud frontier
✈️ Offline Studio Vendored UI; cafe Wi‑Fi optional

One-liner (because of course) 🪄

curl -fsSL https://raw.githubusercontent.com/UnicoLab/smlcode/main/scripts/install-remote.sh | bash

irm https://raw.githubusercontent.com/UnicoLab/smlcode/main/scripts/install.ps1 | iex

brew install --formula https://raw.githubusercontent.com/UnicoLab/smlcode/main/Formula/slmcode.rb

Then: slmcode doctor → Quick start. If doctor is green, you’re dangerous. 💚

Pick your adventure 🎮

🖥️ Terminal vibes

TUI & chat — board, live events, slash commands.

🎨 Clicky cockpit

Studio — kanban + memory, airplane-mode friendly.

🧪 Copy-paste wins

Recipes — workflows that survive reality.

❓ Things broke

FAQ — the “it’s the context window” desk.

☀️ Made with ♥ by UnicoLab · Summer coding with SLMs should feel like a superpower, not a compromise. Now go make doctor green. 💚