claude-almanac 0.3.5

Daily intelligence layer for Claude Code

claude-almanac

A daily intelligence layer for Claude Code — remembers what you've done, summarizes what changed, and surfaces it when you need it.

claude-almanac is a Claude Code plugin that adds three tightly-integrated subsystems to your local development environment:

• Self-curating memory — an LLM picks what's worth remembering from each session and writes structured markdown, scoped per-repo (worktree-safe) and globally.
• Daily digest + Q&A — a background job indexes commits and generates a daily digest per repo, served at http://127.0.0.1:8787 with a local Q&A endpoint.
• Code-index retrieval — a per-repo vector index of public symbols, auto-surfaced alongside memory hits when a prompt looks like a code question.

All three share one pluggable embedder (Ollama by default, OpenAI/Voyage as extras), one XDG-compliant data dir, and one cross-platform install (macOS launchd + Linux systemd).

What makes it different

• Positioning is not "memory for LLMs." The namespace is saturated. claude-almanac leads with daily surfacing — commits, digests, and cross-artifact Q&A — and treats memory as one retrieval source among several.
• Local-first by default. Ollama + bge-m3 + SQLite + local FastAPI server. No cloud dependency in the default path; cloud embedders (OpenAI, Voyage) are opt-in extras.
• Per-repo worktree-safety. Memory and code-index databases key off git-common-dir so worktrees of the same repo share state without colliding.
• Trust boundaries are explicit. The code-index's LLM-powered arch summaries refuse to run unless BOTH the repo's .claude/code-index.yaml AND the global config.yaml opt in via send_code_to_llm: true.

Quickstart (3 steps)

# 1. Install the Claude Code plugin
/plugin install claude-almanac

# 2. Run first-time setup (creates dirs, writes default config, installs platform units)
claude-almanac setup

# 3. Use it
/recall search "authentication"          # past decisions + context
/digest today                            # today's activity digest (opens browser)
/recall code "jwt verification flow"     # per-repo code-symbol search

See docs/install.md for the full per-platform walkthrough, including Ollama install and optional cloud-embedder setup.

Features

Subsystem	What it does	Commands
Memory	Semantic archive of past sessions + curated markdown. Auto-injected at prompt time.	/recall search, /recall search-all, /recall list, /recall show
Digest	Daily per-repo markdown digests + local Q&A web UI.	/digest today, /digest YYYY-MM-DD, /digest generate
Code index	Per-repo vector index of public symbols + optional LLM arch summaries.	/recall code, claude-almanac codeindex init|refresh|arch|status

Supported platforms

Platform	Status	Scheduler	Notifier
macOS 14+	supported	launchd	terminal-notifier / osascript
Linux (Ubuntu 22.04+, Arch, Fedora)	supported	systemd --user	notify-send
Windows	not supported in v0.1	—	—

Python 3.11+ required.

Documentation

• Install — per-platform setup, Ollama install, cloud-embedder setup, troubleshooting
• Config reference — full config.yaml schema, env var overrides, embedder profiles
• Architecture — system map, hooks flow, where to look to change X
• Code index — per-repo symbol indexing, language support matrix, trust boundary
• Contributing — dev setup, adding embedders/platform adapters, test conventions
• Roadmap — what's planned for v0.2 → v0.5+ and how the plugin system will work

License

MIT — see LICENSE.

Status

v0.1.0 — first public release. Feedback via GitHub issues. No telemetry.