Turn feedback into permanent improvements. No retraining. No prompts. Just smarter AI.
Free · Open source · 60 seconds to install
SmartAssist creates a reinforcement learning loop that turns every correction into permanent improvement.
Deep analysis creates architect-level lessons from your codebase
Dual-memory: project rules + past corrections on every prompt
Give :) or :( — per-lesson Thompson Sampling learns
Proven lessons rank higher, bad lessons sink, new ones explore
Your AI gets it right the first time — loop repeats
A complete system that captures, stores, ranks, and retrieves project-specific knowledge.
BGE-M3 + LanceDB hybrid search with cross-encoder reranking
Per-lesson Thompson Sampling — proven lessons rank higher, new ones get fair exploration, bad ones sink automatically
5 lifecycle hooks inject knowledge on prompt, session start, and tool use
9 MCP tools to search, inspect, compare, and evolve lessons from any conversation
MemAlign pattern: injects project rules (semantic memory) alongside past corrections (episodic memory) for richer context
Isolated lessons, vectors, and analytics per project
Analyzes git history, PR reviews, configs, and code structure to create 30-80 lessons automatically — scaled to your codebase complexity
Claude Code, Codex, Gemini, ChatGPT, Amp, OpenCode — Claude is the most deeply validated today, and setup-agent all wires the rest.
Nine MCP tools let you and your AI manage lessons directly from any conversation.
The supported path right now is the GitHub source install. Package registries and one-line installer endpoints are not published yet.
$ pipx install git+https://github.com/jnrahme/SmartAssist.git
$ smartassist setup-agent codex
Use smartassist setup for Claude Code, or smartassist setup-agent all to print/apply every supported agent adapter.
These install channels are still planned:
pipx install smartassistnpm install -g smartassist-memorybrew install smartassist-memory/tap/smartassistcurl -fsSL https://smartassist-memory.com/install | shUse the GitHub install tab today. This panel will be updated when a package channel is actually live.
$ git clone https://github.com/jnrahme/SmartAssist.git
$ cd SmartAssist && pipx install .
Best for local development and testing SmartAssist from a checkout before package channels are published.
$ pipx install git+https://github.com/jnrahme/SmartAssist.git
$ cd my-project && smartassist setup
This is the supported install path today. For extra agents after install, use smartassist setup-agent <agent>.
Supported today: GitHub source install · Python 3.10+ · Project-scoped setup for Claude Code plus adapter setup for other agents
Register with any AI agent
All agents get the same MCP tools and learning loop. Claude Code gets the richest experience with automatic hook injection on every prompt.
Install SmartAssist and make every project smarter from day one.
Install SmartAssistBUSL-1.1 licensed · Works with Claude Code, Codex, Gemini, ChatGPT, Amp, and OpenCode