Open Source · MIT · Runs 100% locally

AI that learns things
you didn't teach it

Familiar is an AI companion that lives on your machine. It installs alongside OpenCode, trains local models on every session, and develops knowledge you never explicitly gave it.

The more you use it, the more it knows. Your data never leaves your machine.

OpenCode Ollama Forge™ MCP Bun TypeScript
Copied! $ curl -fsSL familiar.run/install.sh | bash

Installs Familiar + OpenCode + Ollama models · macOS / Linux

opencode — ~/my-project — powered by familiar
opencode v1.2 · familiar/qwen2.5:7b-instruct MCP: familiar (14 tools) · LSP: typescript · Forge: 47 pairs today
you >
the login form doesn't validate email format, can you fix it?
[glob] src/components/Login*.tsx
[read] src/components/LoginForm.tsx
[edit] src/components/LoginForm.tsx +42
familiar
Added email validation to LoginForm.tsx. The handleSubmit function now checks format before calling the API. LSP confirms zero errors. (session logged to Forge)
familiar model: qwen2.5:7b forge: training turn 1 · 847 tokens
> ask anything...

The only AI that learns from you

Other AI tools ship a frozen model. You use it, they get your data, the model never changes. Familiar is different.

The Forge™ pipeline captures every session as training data, fine-tunes your local models, and deploys them back — automatically. The more you work, the more it understands. Not just code — your patterns, your decisions, your thinking. Your data stays on your machine. Your models improve for you.

Other tools

  • Static models, same for everyone
  • Cloud API dependency
  • Your data sent to third parties
  • Generic, one-size-fits-all
  • No memory between sessions

Familiar

  • Models that learn your patterns
  • Runs 100% locally via Ollama
  • Your data never leaves your machine
  • Adapts to your style and thinking
  • Persistent memory across all platforms

How Familiar works

Familiar is the brain. OpenCode is the hands. Together they form a self-improving system that runs entirely on your machine.

Familiar

Always-on AI companion that remembers everything, trains on every session, and develops knowledge you didn't give it.

~ Persistent memory (SQLite + FTS5)
~ Multi-model routing
~ Forge training pipeline
~ Activity sync across platforms
~ Autonomous background agents
~ Scheduled tasks (cron)

OpenCode coding

Terminal coding agent by the Charm team. Familiar installs it, configures your local models as the provider, and connects via MCP so every tool call, file edit, and session feeds back into the brain.

Telegram / Slack channels

Get notifications, morning briefs, PR summaries, and training reports. Ask Familiar questions from your phone. All conversations feed back into model training.

MCP Bridge 14 tools

Familiar exposes 14 MCP tools: memory search, observations, system status, smart routing, and more. Any MCP-compatible client can connect to the brain.

All models run locally via Ollama. Swap models in opencode.json. Falls back to cloud APIs when local models are unavailable.

What you get

One install. A companion that grows with you.

OpenCode integration

The installer sets up OpenCode with your local Ollama models as the provider and Familiar as an MCP server. OpenCode handles the coding; Familiar handles the learning.

Self-improving models

Every session generates training data. Forge fine-tunes your local models and deploys better versions. Version 1 was generic. Version 15 knows how you think.

Persistent memory

SQLite + full-text search. Familiar remembers your decisions, blockers, preferences, and context. Search across sessions with familiar_memory_search.

Cross-platform sync

Code in OpenCode, check in from Telegram, review on Slack. All activity syncs through the brain. One unified timeline of everything you're doing.

14 MCP tools

Memory search, observations, chat routing, system status, session management, and more. Any MCP client connects to the full brain. Not just OpenCode.

100% private

Models run on your machine via Ollama. Training happens locally. No data leaves your network. No API keys required for local operation. Your companion, your data.

Forge™: the training loop

This is what makes Familiar alive. Every session feeds the pipeline. Your models develop understanding you never explicitly taught them.

Collect
Every session saved as JSONL training pairs
→
Prepare
Deduplicate, filter, split train/valid
→
Train
Fine-tune with LoRA on your GPU
→
Deploy
Quantize and hot-swap in Ollama
→
Eval
Benchmark against previous versions

Manage with familiar forge status, familiar forge train, familiar forge eval, familiar forge rollback

Get started in 3 steps

1

Install

One command installs Familiar, OpenCode, and wires them together. Pulls recommended Ollama models automatically.

curl -fsSL familiar.run/install.sh | bash
2

Use it

Open any project and run opencode. It's already configured with your local models and connected to Familiar.

cd my-project && opencode
3

Watch it learn

After a few sessions, run Forge to train. Your models improve. Keep working. Keep training. It never stops learning.

familiar forge train

Familiar is built from scratch. No LangChain, no AutoGen, no third-party agent frameworks. The training pipeline, memory system, activity sync, MCP bridge, and model serving layer are all original code by engindearing. OpenCode handles the coding TUI because it's already the best terminal agent out there. We focus on making your AI smarter.