Repository navigation
Releases
Palmi v26.11
Download Palmi on the App Store
What’s New in v26.11
Background Processing
- Fixed background processing. After you leave Palmi or lock your device, tasks keep running within the time the system allows, with live progress on the Lock Screen and in the Dynamic Island.
- Background processing is now on by default and can be turned off in Settings.
Interactive Replies
- Replies can include native charts with sliders, toggles and calculators, as well as small HTML apps such as games, animations and simulations. No extra setup is required.
- Replies appear as they are written, keeping text, images and interactive content in order. Suggested next messages help continue the conversation.
Built-in Browser (Beta)
- In Professional mode, direct the built-in browser in everyday language. Palmi opens pages, clicks, types and moves between tabs.
- When a page asks for a sign-in or verification, take over and complete it yourself.
Model Library
- Models are organized by provider, with one default model in place of the previous primary, light and multimodal plan. Existing configurations migrate automatically, and a one-page summary shows how the previous setup maps to the new one. The summary can be reopened at any time from Model Settings.
- Switch models within a conversation. Automatic approval, vision for models that cannot see images, remote web search and page reading, and image generation can each use a separate model.
Unified Chat Shell
- Palmi opens straight into a conversation, with a consistent experience across Chat and Professional modes.
- A new sidebar brings conversation history, files, skills and settings together. Unread counts appear as numbers on the sidebar button and on each mode.
- Browse project files together, open folders, preview and share files, and view conversation attachments more easily.
- When Palmi needs a decision, the question appears over the current screen. Answer in sequence, add details, or return to the conversation that asked.
- Dark Mode is now supported. Palmi follows the system appearance.
Long-Running Tasks
- Background work reports progress more promptly, including Bionic replies.
- Chat and Professional tasks are no longer interrupted by the previous fixed time and turn limits.
Languages and Privacy
- Added German, Swedish, Arabic, Turkish, Indonesian and French. Palmi is now available in 11 interface languages, and replies and conversation titles follow the chosen language more closely.
- First launch now asks you to accept two documents separately: the Third-Party AI Service Data Notice and the Privacy Policy. Both are provided as PDFs in all 11 languages.
Fixes and Changes
- Fixed losing the current conversation after collapsing the browser.
- Fixed the wrong mode being selected when opening the sidebar from Bionic mode.
- Improved read status and unread reminders in Bionic conversations.
- Removed the prompt that suggested switching to Bionic mode while a Professional task was running.
Palmi uses the model services you configure. Neither the app nor Bionic Pro includes model-service credits.
Repository: Hyp6666/PalmiAgent · Full changelog: v26.10...v26.11
Palmi v26.10
Download Palmi on the App Store
What’s New in v26.10
Bionic Mode
- Added virtual character conversations with character profiles, ongoing memories, fictional diaries, and proactive messages.
- Create characters manually, import character archives, or use the built-in character-creation skill and tools in Professional mode.
- Added conversation search, immediate or natural reply timing, pinned chats, muted notifications, custom chat backgrounds, and archive export and import.
- Bionic characters can delegate focused research and analysis tasks to Professional mode and use the returned results in the conversation.
- Bionic Basic includes the built-in Palmi and up to two custom characters. The one-time Bionic Pro unlock raises the custom-character limit to 99 and adds memory details, source-message navigation, manual memory editing, diary reading, and the developer inspection panel. Both plans use the model services you configure; neither includes model-service credits.
Codex OAuth and Image Generation
- Added Codex OAuth sign-in, account-associated model discovery, and import into the global model library.
- Added dedicated Image Generation Settings and image-generation tools, including images in Bionic conversations. Model availability, image access, and usage allowances depend on the connected account.
Neo Display Style
- Added Neo as a new display style for the existing task process in Chat and Professional modes. It presents thinking, progress messages, and tool activity in a timeline with expandable steps and elapsed time.
- Neo is now the default display style. Switch between Neo and Hardcore in runtime settings; your selected style is saved across launches.
Conversation and Settings Improvements
- Added unread indicators across modes and conversations, with persistent read state and notification badge synchronization.
- Added custom response-style descriptions of up to 200 characters and separated description expansion from style selection.
- Refined model selection, message presentation, and the Bionic plan comparison interface. Updated explanations are available in all five interface languages.
PalmiAgent v26.9
Download PalmiAgent on the App Store
What’s New in v26.9
More Ways to Connect Your Models
- Added protocol selection to the Add Model and model editing screens. Choose OpenAI Chat Completions, OpenAI Responses, or Anthropic Messages to connect your preferred service.
- Added automatic protocol matching to help connect compatible model services with less manual setup.
Local and Remote Web Search
- Added Search Configuration in Settings, where you can switch between local and remote web search.
- Local search now lets you choose one source: Baidu, Bing, DuckDuckGo, Sogou, or 360 Search.
- Added remote search configurations for services with server-side web search. Save your service address, model, and API key, validate the connection, and switch between saved configurations. Responses and Messages services are supported.
PalmiAgent v26.8
Download PalmiAgent on the App Store
PalmiAgent is a local-first, model-flexible personal AI agent workspace for iPhone. It can plan multi-step work, use tools, run Python, research the web, understand files and images, and keep every result together in a persistent workspace.
Version 26.8 makes technical answers substantially easier to read, inspect, and reuse.
What’s New in v26.8
Native Math Rendering
- Inline and display LaTeX now render directly inside conversations.
- Common delimiters and
math,latex, andtexcode fences are supported. - Copying rendered formulas preserves the original LaTeX source.
Better Code Blocks
- Fenced code blocks now include language-aware syntax highlighting.
- Long lines remain readable with horizontal scrolling.
- Each block has a clear language label and one-tap copy action.
- Incomplete code fences remain stable while a model is still streaming its response.
More Consistent Model Output
- PalmiAgent now gives models a shared formatting contract for Markdown, code, math, links, and tables.
- Technical responses are more predictable across providers and easier to scan on a phone.
A New Product README
- The English README is now the default and presents PalmiAgent as a complete product, not only a source repository.
- A full Simplified Chinese edition is included.
- Both editions use nine product screenshots in a natural feature-by-feature narrative, with localized screenshots for Chinese readers.
Release Quality
- Updated dependency metadata and third-party license notices for the native math renderer.
- Added Xcode Cloud project metadata.
- Strengthened request-compatibility tests for current Foundation behavior and refreshed stale runtime coverage.
Requires iOS 26.1 or later. PalmiAgent does not include model weights or third-party model credits; connect a compatible cloud or local model service to get started.
PalmiAgent v26.7
Download PalmiAgent on the App Store
PalmiAgent is an AI workspace for iPhone that brings conversations, project files, models, images, web research, Python execution, and system tools into one place. It is designed to preserve context and help AI move real work forward—not just answer isolated questions.
Built for Real Work
- Persistent workspaces: Keep conversations, attachments, generated files, and tool results together inside projects and threads.
- Flexible model plans: Combine primary, multimodal, and lightweight models from GLM / Z.AI, DeepSeek, local services, or custom OpenAI-compatible endpoints.
- Multimodal understanding: Add images from Camera, Photos, or Files, then use a vision model or local OCR to analyze them.
- Research and execution: Read webpages and documents, work with project files, and use the embedded Python environment for calculations, data processing, and automation.
- Clear user control: Configure your own API keys and choose how PalmiAgent authorizes tools that access data, modify files, or open system features.
What's New in v26.7
- A redesigned first-run experience for choosing your language, AI service, and Chat or Pro mode.
- Interface support for Simplified Chinese, Traditional Chinese, English, Japanese, and Korean.
- Improved Camera, Photos, Files, webpage, and long-document workflows.
- Conversation usage indicators plus automatic and manual context organization in Pro mode.
- Clearer progress states, response actions, link handling, confirmation prompts, and result presentation.
- More consistent settings pages, expanded privacy information, and a smoother reset and reconfiguration flow.
PalmiAgent v1.0.1
Changes
- Added a grouped settings center for model, tools/knowledge, experience, and system sections.
- Added system data management, system info, and privacy/policy entry points.
- Refactored chat/pro agent prompt context into surface-aware runtime layers, with turn-scoped directives kept out of the stable system prompt.
- Improved project/session handling so project folders are independent from thread selection and session state is persisted more cleanly.
- Removed the fixed 300s LLM resource timeout so long streaming/thinking responses are not cut off by the local URLSession resource limit.
- Adjusted chat top chrome spacing and synced the pre-release todo checklist into the repository.
Validation
xcodebuild -project PalmiAgent.xcodeproj -scheme PalmiAgent -configuration Release -sdk iphonesimulator buildsucceeded.
Release status
This is saved as a draft/pre-release record. PalmiPrivacyPolicy.md still contains publication date, operator, address, and contact placeholders, so it should not be treated as a formal public production release until those legal/operational fields are replaced and reviewed.
PalmiAgent v1.0.0
Download PalmiAgent on the App Store
PalmiAgent v1.0.0 brings together flexible model management, multimodal understanding, a refined chat experience, and stronger reliability for long-running work.
Highlights
- Rebuilt model management with clearer provider settings, model roles, and custom OpenAI-compatible configurations.
- Improved setup and validation for GLM / Z.AI, DeepSeek, local models, and custom endpoints.
- Added multimodal workflows with bundled PP-OCRv6 Tiny OCR, image text recognition, configurable vision models, and multi-image input.
- Refined the chat composer, attachment previews, shared controls, and core interface components.
- Expanded planning and goal-oriented workflows for more complex tasks.
- Added streaming reasoning so progress is visible while a model is working.
- Improved background persistence, session recovery, and task state handling for greater reliability.
- Refreshed the app icon, README visuals, and third-party OCR notices.



