spool-memory 0.2.3

Local-first developer memory system — persistent, structured knowledge for AI coding tools
Documentation
# Roadmap

## Current State

`spool` is currently a local-first developer memory toolchain with:

- project-based vault scanning
- heuristic note ranking
- multi-format rendering
- wakeup packet generation and policy-aware rendering
- append-only lifecycle ledger and latest-state projections
- CLI, MCP, GUI, daemon, and desktop-backend surfaces
- experimental GUI behind a feature flag

## Relationship To The Original Plan

The original planning note defined a three-step path:

1. CLI
2. MCP
3. desktop client

That sequence is still valid.

The current roadmap expands the scope around that sequence by adding:

- security and policy enforcement
- memory ledger separation
- wake-up packets
- cross-AI memory continuity

## Phase 0: Harden The Router

Status: mostly complete, still open for retrieval-quality improvements

Goals:

- improve retrieval precision
- add benchmarks
- stabilize config and cache behavior
- document product direction

Deliverables:

- better field-aware ranking
- note indexing or equivalent derived-state support
- scan cache
- retrieval and end-to-end benchmarks
- design docs and ADRs

## Phase 1: Developer Memory Core

Status: complete in minimum viable form

Goals:

- introduce memory object schema
- add wake-up packet generation
- distinguish curated vault notes from captured memory

Deliverables:

- `memory_type` aware retrieval
- source-of-truth boosting
- developer profile packet
- project wake-up packet
- retrieval policy by sensitivity

## Phase 2: Memory Ledger

Status: partially complete

Goals:

- capture raw memory events locally
- avoid polluting curated Obsidian notes

Deliverables:

- local append-only session ledger
- importers for transcript, git, and terminal metadata
- memory extraction pipeline for decision and incident candidates
- promotion flow from ledger to curated notes

## Phase 3: Cross-AI Gateway

Status: complete in minimum viable form

Goals:

- support multiple AI clients through one local interface

Deliverables:

- MCP server
- stable CLI subcommands for search, wake-up, and explain
- local daemon mode for optional read-side reuse
- client-specific policy shaping

## Phase 4: Retrieval Intelligence

Status: not started

Goals:

- increase precision and long-term maintainability

Deliverables:

- stale-memory detection
- contradiction detection
- confidence scoring
- semantic retrieval with local embeddings
- better excerpt extraction

## Phase 5: Optional Compact Memory

Status: not started

Goals:

- reduce context cost without sacrificing trust

Deliverables:

- layered summaries
- room digests
- dense wake-up packets
- optional compact memory representation

This phase should happen only after structure, provenance, and security are stable.

## Immediate Next Milestones

### Milestone A

Sync roadmap/product docs with the implemented memory lifecycle baseline so restart decisions match current code reality.

### Milestone B

Decide the next major build direction:

- harden and extend the new minimal desktop host on top of the existing desktop facade/Tauri adapter boundary, or
- move directly into retrieval intelligence work

### Milestone C

If desktop comes first:

- extend the new runnable desktop host beyond its initial bootstrap slice
- wire a minimal frontend shell to existing context / wakeup / lifecycle flows

### Milestone D

If core quality comes first:

- improve excerpt selection
- expose clearer score diagnostics
- evaluate heavier retrieval index work before semantic retrieval

### Milestone E

Resume unfinished Phase 2 work:

- transcript / git / terminal importers
- extraction pipeline for decision and incident candidates
- promotion flow from ledger to curated notes

## Decision Discipline

Any roadmap change that affects:

- product scope
- security guarantees
- memory storage model
- integration surface

should be written as an ADR in `docs/adr/`.