Skip to content

README.md

Latest commit

 

History

History
47 lines (29 loc) · 1.86 KB

README.md

File metadata and controls

47 lines (29 loc) · 1.86 KB

memup

Self-learning memory system for AI coding agents (Claude Code + Codex CLI).

Foundation milestone (plan 1 of 5). This is a minimal CLI that can save and recall memories from a local SQLite store. Hooks, daemon, hybrid search, and the self-learning loop arrive in plans 2–5. See docs/superpowers/specs/2026-05-03-memup-self-learning-memory-design.md.

Build

cargo build --release

The binary lands at target/release/memup.

Prerequisites

memup currently expects an explicit path to the BAAI/bge-small-en-v1.5 ONNX model and its tokenizer files (later plans bundle this).

Download the model and tokenizer files into one directory, then point at the ONNX file:

export MEMUP_MODEL_PATH=/path/to/bge-small-en-v1.5/model.onnx

Required sibling files in that same directory:

  • tokenizer.json
  • config.json
  • special_tokens_map.json
  • tokenizer_config.json

(Files come from https://huggingface.co/BAAI/bge-small-en-v1.5/tree/main. Download once on a connected machine; copy into the air-gapped environment.)

Quick smoke

target/release/memup init
target/release/memup save "We chose JWT over session cookies for stateless edge auth." --kind decision
target/release/memup save "Redis cache TTL is 5 minutes for /orders." --kind fact --tags caching,redis
target/release/memup recall "auth tradeoff"
target/release/memup recall "cache duration" --k 3 --json
target/release/memup doctor

Per-project memory

By default, memup looks for a .git directory walking up from CWD. If found, the store lives at <repo>/.memup/local/index.db. Otherwise it falls back to ~/.memup/projects/default/index.db. Override with:

export MEMUP_HOME=/some/where

Tests

cargo test --workspace                                # unit + integration (fast)
MEMUP_TEST_MODEL_PATH=$MEMUP_MODEL_PATH cargo test    # also runs embedder/CLI e2e