1.5.0

Changelog

1.5.0

Features

• Add show/hide message icons toggle to appearance settings (d7cbc20)
• Fix copy-to-clipboard for code blocks and fix code block overflow (fc2e1e3)
• Fix chat header truncation and shrink button group (612bf7d)

Fixes

• Fix Windows 1.4.0 build (db28afd)

1.4.0

Changelog

1.4.0

Features

• Add PDF attachment text extraction so PDF attachments are sent to models as extracted text (35280dd)
• Give the chat area more room and adjust the native menu; fix provider list sync (00a8852)

1.3.0

Changelog

1.3.0

Features

• Add context token estimator with a live context-usage indicator in the message input footer (adcad13, 14f67fa)
• Persist per-conversation summaries with a summarized_up_to_id cutoff via a new schema migration (974b43a)
• Implement automatic context compaction: summarize older turns once a conversation approaches its model's context window, keeping a raw tail unsummarized (9f0b590)
• Add a manual "Compact" button and resolve real context windows from provider model metadata (b8bd005)
• Invalidate a conversation's persisted summary automatically when an edit or regenerate touches the already-summarized region (7953788)

1.2.0

Changelog

1.2.0

Features

• Add Google and DuckDuckGo search window tools that open dedicated webview windows for manual searching, with result scraping and fetch chaining (8a7c794)
  • Injects a floating "Done - send results" button into both search engines for collecting visible result links
  • Handles Google redirect URLs and DuckDuckGo uddg redirect URLs transparently
  • Falls back to returning visible page text when no result links can be parsed
  • Includes settings toggles for both search providers in the Web Search settings tab
• Update fetch-url tool description to guide agents to chain fetches after search window results

1.1.0

Changelog

1.1.0

Features

• Add web search and URL fetch tools, with provider settings UI for configuring web search (59816c5)
• Add fixed/fluid chat width layout option with new appearance settings (7e4d382)

Fixes

• Fix message list bug and update fetch URL handling (798db5b)

Improvements

• Improve provider sync flow and polish settings UI (481b458)

Documentation

• Add architecture documentation (ARCHITECTURE.md), remove unused greet command (1ddb626)
• Update README links and add architecture doc reference (af820e6, 8173d4d)

1.0.0

• Initial release.

v1.0.0

v1.0.0 - Initial Release

Personal Agent is a local-first AI chat desktop app that works with any OpenAI-compatible provider. Define custom instructions, skills, and agents to get exactly the responses you want - no generic outputs, no unwanted emojis.

What You Can Do

Chat with any provider

• OpenAI, DeepSeek, Opencode Go, Ollama, LM Studio, or any OpenAI-compatible endpoint
• Streaming responses with collapsible thinking/reasoning content
• Edit messages, retry on errors, regenerate responses
• Keyboard shortcuts: Enter to send, Shift+Enter for newline, Esc to stop

Customize how the AI responds

• Custom instructions - global rules injected into every conversation (no emojis, specific formatting, preferred response structure)
• Skills - reusable knowledge blocks you activate per conversation (coding conventions, project guidelines, domain rules)
• Custom agents - combine instructions and skills into named profiles for different workflows

Organize your conversations

• Tag and pin conversations
• Search across all conversations with debounced input
• Auto-generated conversation titles
• Virtualized message list for smooth scrolling in long chats

Sync across devices

• Optional Turso database for cloud persistence
• Conversations, agents, skills, and instructions follow you across macOS and Windows

Privacy-first

• All data stored locally by default
• API keys masked in all UI and never logged
• Export provider settings without keys

Tech Stack

Tauri v2, React 19, TypeScript, Zustand, Vercel AI SDK, Tailwind CSS v4, shadcn/ui, Turso (libsql)