Pi Agent

Chat with Pi in Obsidian using context from your current note, links, backlinks, tags, explicit search results, and selected text.

Thanks to Mario Zechner, the developer of Pi, for building the agent this plugin runs on top of.

Requirements

Pi Agent is desktop-only and requires the Pi coding agent to be installed separately. The upstream Pi coding agent package is @earendil-works/pi-coding-agent, from earendil-works/pi-mono:

npm install -g @earendil-works/pi-coding-agent
pi --version

If Obsidian cannot find pi, restart Obsidian after installation so it picks up your updated PATH.

First run checklist:

1. Install and authenticate Pi from a terminal.
2. Open a dedicated test vault before enabling Edit or Full agent mode.
3. Start in Chat or Review mode until you understand what context is sent.
4. Enable Edit or Full agent only for vaults/projects you are comfortable letting Pi modify.

Tool modes, briefly:

• Chat attaches Obsidian context only; Pi CLI tools are disabled.
• Review lets Pi read/search/list files.
• Edit lets Pi edit/write files.
• Full agent also lets Pi run shell commands.

Privacy reminder: prompts, selected text, note content, search excerpts, attachments, and local chat history can be sent to the Pi CLI and then to your configured model provider.

Features

• Chat with Pi from an Obsidian sidebar.
• Attach current-note context automatically.
• Include linked notes, backlinks, tags, frontmatter, headings, selected text, and explicit search attachments.
• Choose tool modes: Chat, Review, Edit, and Full agent.
• Enable default Pi skills and add trusted custom skill folders.
• Use / autocomplete for Obsidian context commands and /skill:name commands.
• Review detected vault changes and diffs after edit/full-agent runs.
• Copy responses, create notes from answers, and open cited vault notes.

Tool modes control which Pi CLI tools are enabled. They are not an operating-system sandbox.

Privacy and safety

Pi Agent can send note content and selected text to the local Pi CLI, which may forward prompts to configured model providers. See PRIVACY.md for details before publishing or using the plugin with sensitive vaults.

Short version:

• The plugin does not include ads, telemetry, or an auto-updater.
• Chat history and Pi sessions are stored locally by the plugin and Pi.
• Network access happens through the separately installed Pi CLI and depends on your Pi/model-provider configuration.
• The plugin reads Pi configuration and skill files from vault/project .pi/ and .agents/ folders, plus any absolute or vault-relative skill folders you configure.
• Edit and Full agent modes can modify files in your vault/project.
• Full agent mode can run shell commands through Pi.
• Skills can contain instructions or scripts; only enable skill folders you trust.

Installation

Community plugins

After approval, install from Obsidian's Community Plugins browser.

Manual installation

Download the latest release and copy these files into:

<vault>/.obsidian/plugins/pi-agent/

Required files:

main.js
manifest.json
styles.css

Then enable Pi Agent in Obsidian settings.

Development

Use a dedicated test vault. Do not develop or test plugin changes in your main vault.

npm ci
npm run build
npm run build:check
npm run ci
npm run dev:install -- /path/to/vault/.obsidian/plugins/pi-agent

Then reload Obsidian, or disable and re-enable the plugin.

See docs/development.md and docs/architecture.md.

Release

1. Create a release-prep branch from main.
2. Update manifest.json, package.json, and versions.json; promote CHANGELOG.md Unreleased entries into the new version section.
3. Run:

npm run ci

4. Commit and merge the release prep into main.
5. Create and push a matching SemVer tag from main, for example:

git tag 0.0.1
git push origin 0.0.1

The release workflow uses the current CHANGELOG.md entry as release notes, publishes the Obsidian-supported assets, and generates artifact attestations:

• main.js
• manifest.json
• styles.css