AgentVis v0.3.5

AgentVis v0.3.5


This release adds a native conversation history recall tool for Agents, improving long-running continuity after context compression.


Added



• Added conversation_search, a read-only native tool that lets an Agent retrieve its own historical conversations.
• Supports three recall modes:
• timeline: browse message history by time without needing a keyword.
• search: keyword search with lightweight centered snippets.
• get: fetch full selected historical messages by message ID.
• Added pagination with hasMore / nextOffset for robust recall across large histories.
• Added time-range filtering for locating conversations from a specific day, week, month, or period.
• Results are scoped to the current Agent only, preventing cross-Agent history mixing.
  

Improved



• Master Brain guidance now encourages dispatching a Sub-Agent to retrieve past conversations when memory, RAG, or compressed chat history is insufficient.
• Conversation search uses memory-safe historical text so snippets and fetched messages align with what the Agent can safely reason over.
• Added native skill conflict protection for newly added native tools.

AgentVis v0.3.4

AgentVis v0.3.4

This is the first public release of AgentVis, a local-first AI agent workspace for Windows.

Highlights

• Visual agent runtime with Master Brain and Sub-Agent collaboration.
• Local desktop workspace powered by Tauri, React, TypeScript, and Rust.
• Built-in support for memory, skills, RAG knowledge bases, scheduled tasks, visual artifacts, and human-in-the-loop intervention.
• Security-oriented execution model with tool interception, command validation, sandbox auditing, and Trash Bin soft delete.
• Feishu / Slack remote control support for triggering and monitoring agent tasks.

Download

Download the Windows installer from the assets below:

• AgentVis_0.3.4_x64-setup.exe

Notes

• Current release is Windows-first.
• Some sandboxing, credential, WFP, desktop automation, and runtime capabilities are designed primarily for Windows.
• This is an early open-source release; feedback, issues, and contributions are welcome.