Recall

A local context engine for AI-assisted development. Recall maintains persistent, searchable knowledge about your projects and coding patterns, reducing redundant context building and token consumption across AI assistant sessions.

What It Does

Recall operates as a context layer between your codebase and AI assistants:

• Indexes project structure, file summaries, and semantic content into local SQLite databases
• Matches user prompts against reusable task recipes and historical solutions
• Routes simple queries to local LLMs (llama.app, llama.cpp, Ollama) and complex ones to cloud assistants
• Learns from interactions: tracks what local models handle well, extracts reusable code snippets, records agent behavior

Nothing leaves your machine unless you delegate to a cloud model.

Installation

# macOS / Linux
curl -LsSf https://llama.app/install.sh | sh  # optional: for local LLM support
go install github.com/gleicon/recall@latest

Recall is a single static binary. No runtime dependencies, no CGO, no API keys required.

Quick Start

# 1. Map your project
cd my-project
recall map

# 2. Load default recipes (framework patterns, common tasks)
recall recipes seed

# 3. Generate a context-rich brief for an AI assistant
recall brief "add OAuth login"

# 4. Query with smart routing (local model if available, else enriched brief)
recall query "how do I structure middleware in this project"

Core Commands

Project Context

Command	Purpose
recall map	Detect and cache project type, entry points, module boundaries
recall cache build	Index all source files with summaries and embeddings
recall cache inspect	View cached files, subsystems, and memories
recall cache refresh	Incrementally update only changed files
recall learn "insight"	Store a manual insight into project memory

Knowledge Retrieval

Command	Purpose
recall brief "task"	Generate enriched prompt context from recipes + project state
recall query "question"	Smart router: local LLM answer or delegation brief
recall search "terms"	FTS5 + vector search over indexed content
recall search -c "terms"	Chunk-level semantic search

Global Brain

Command	Purpose
recall brain conversations	History of local model interactions
recall brain snippets	Reusable code blocks extracted from responses
recall brain lessons	What works per framework / model
recall brain stats	Aggregate metrics: success rate, tokens saved
recall brain search "auth"	Search all brain data
recall brain frameworks	Per-framework performance breakdown

Recipes

Command	Purpose
recall recipes seed	Load 30+ default recipes (Go, Next.js, Python, Rust, etc.)
recall recipes list	Show all loaded recipes with usage counts
recall recipes add -f my-recipe.json	Add a custom recipe

Local LLM Management

Command	Purpose
recall local status	Detect running local LLM server
recall local models	List available models
recall local use <model>	Lock to a specific model

Recording & Stats

Command	Purpose
recall run suggest --task "..."	Gated recording of an assistant run
recall run record --task "..."	Manual recording without prompt
recall stats cache	Project cache statistics
recall stats recipes	Recipe usage statistics
recall stats runs	Aggregated run statistics
recall stats global	Cross-project global statistics
recall stats insights	Most/least useful recipes

Feedback & Quality

Command	Purpose
recall feedback --good	Mark last query answer as accepted
recall feedback --bad --note "reason"	Mark last query answer as rejected

Tool Integration

Command	Purpose
recall --version	Print binary version
recall status	JSON: version, mapped, mapped_at, language, framework, files_indexed

recall status is designed for scripts and editor plugins to detect whether recall is installed and a project has been bootstrapped:

recall status
# {"version":"v0.3.1","mapped":true,"mapped_at":"2026-06-05 10:00:00","language":"go","framework":"go-module","files_indexed":87}

Maintenance

Command	Purpose
recall bench	Run performance benchmarks
recall cleanup	Remove old cache entries
recall cleanup project <dir>	Remove a project's data directory

Data Storage

Recall uses two SQLite databases:

Global (~/.recall/global.db):

• Task recipes with vector embeddings
• Framework fingerprints
• Conversation history with local models
• Extracted code snippets
• Agent lessons (what works per framework/model)

Per-project (~/.recall/projects/<hash>/project.db):

• File summaries and content
• Semantic chunks with embeddings
• Subsystem abstractions
• User insights and memories
• Run history and outcomes

All data is local. No cloud sync, no telemetry.

Documentation

• Usage Reference — Complete command reference with examples
• Integration Guide — Hook into Claude, Pi, opencode, Cursor, Codex
• Scripting & Automation — Shell integration, pre-commit hooks, CI
• Manual Data Management — Adding recipes, insights, snippets by hand
• Backup & Migration — Export, import, and archive strategies
• Architecture — How the embedding, search, and routing systems work

Requirements

• Go 1.22+ (for building from source)
• macOS or Linux
• Optional: Any OpenAI-compatible local LLM server on localhost:8080

License

MIT