Turn local AI coding sessions into a searchable Obsidian knowledge base.
llm-iwiki collects sessions from Claude Code, Cursor, Codex, Gemini, and CodeBuddy, normalizes them by project in SQLite, prepares AI-friendly summarization tasks, stores accepted project learnings, and exports navigable Markdown notes to Obsidian.
- Collect local AI coding sessions across multiple tools.
- Resolve sessions by project, repository, path, or slug.
- Generate summarization and experience-extraction task files for AI assistants.
- Review and accept reusable engineering experiences before publishing them.
- Export project summaries, session notes, and project-scoped experiences to Obsidian.
- Preserve user-written note sections while updating managed content blocks.
- Search session summaries and accepted experiences from the CLI.
- Node.js 20 or newer at runtime.
- Bun for local development, testing, and building.
SQLite is provided by better-sqlite3; prebuilt binaries are installed automatically by npm where available.
npm install -g @codehourra/llm-iwikiYou can also run it without installing globally:
npx @codehourra/llm-iwiki --helpllm-iwiki init
llm-iwiki sync
llm-iwiki projects list
llm-iwiki projects inspect .Configure an Obsidian vault and export notes:
llm-iwiki config set obsidian.vault ~/Obsidian/Vault
llm-iwiki summarize prepare changed --project .
llm-iwiki summarize apply --project . --file .llm-iwiki/exchange/summaries.yaml
llm-iwiki experiences prepare --project . --from changed-summaries
llm-iwiki experiences propose --project . --file .llm-iwiki/exchange/experiences.yaml
llm-iwiki experiences candidates --project .
llm-iwiki experiences accept <candidate-id>
llm-iwiki obsidian export --project .The prepare commands generate task files for an AI assistant. Write the corresponding YAML files from those task files, then apply/propose them with the commands above.
llm-iwiki sync [--project <path>]
llm-iwiki projects list
llm-iwiki projects inspect <project>
llm-iwiki projects dedupe
llm-iwiki sessions list --project <project>
llm-iwiki sessions read <session-id>
llm-iwiki summarize prepare [changed|all] --project <project>
llm-iwiki summarize apply --project <project> --file <summaries.yaml>
llm-iwiki experiences prepare --project <project> [--since 30d]
llm-iwiki experiences propose --project <project> --file <experiences.yaml>
llm-iwiki experiences accept <candidate-id>
llm-iwiki search <sessions|experiences> <query>
llm-iwiki obsidian export [--project <project>|--all]
llm-iwiki obsidian check<project> can be a local path, a proj_xxx id, a display name, or a slug. If multiple projects match, use projects list to pick the exact project id.
Exports are written under LLM-iWiki/ in your configured vault:
LLM-iWiki/
├── README.md
├── Projects/
│ └── <project-slug>/
│ ├── Project Summary.md
│ ├── sessions/
│ └── experiences/
└── Topics/
Generated notes use frontmatter plus an aiwiki:managed block. You can write your own notes outside the managed block; re-exporting preserves those sections unless --force is used.
Install the bundled aiwiki-knowledge skill into the current project:
llm-iwiki skills init [--target codex|claude-code|cursor] [--force] [--dry-run]The skill teaches compatible assistants how to run the sync -> summarize -> experiences -> export workflow and how to produce valid YAML for summarize apply and experiences propose.
bun install
bun run typecheck
bun test
bun run buildThis project uses Changesets for versioning, changelog generation, and npm publishing:
bun run changeset
bun run version-packages
bun run releaseUser-facing changes should include a .changeset/*.md file. GitHub Actions creates the release PR and publishes to npm after the release PR is merged.
See CHANGELOG.md.
See docs/roadmap.md.