MS Agent

Open-source Python AI agent framework with a local browser workspace, project memory, multi-agent workflows, and OpenAI, Anthropic and ModelScope support.

Screenshot of MS Agent website

MS Agent is an open-source Python framework for developers building AI agents that carry out complex tasks over multiple sessions. Its browser workspace works with local projects, while terminal interfaces and a Python SDK share the same agent logic. It uses the Apache 2.0 license and supports OpenAI-compatible APIs, Anthropic and ModelScope.

The framework combines models, tools and reusable skills with controls for planning, permissions and execution feedback. Developers can add business rules and decide when an agent needs human intervention. The WebUI lets users chat with models, inspect tool activity, manage skills and MCP tools, and browse or edit project files.

Long-running work gets separate session history, active context and project memory. Agents can retain full records while shortening older context and trimming tool output, then reuse project knowledge in later tasks. Scheduling supports recurring and one-time jobs.

A lead agent can delegate work to specialists, or developers can define workflows with explicit stages and dependencies. Built-in applications include Agentic Insight for research reports tied to evidence, CodeGenesis for software projects checked with diagnostics and runtime feedback, and FinResearch for analysis using AkShare, BaoStock and web information.

DocResearch reads PDF, TXT, PPT and DOCX files and exports illustrated reports. Singularity Cinema coordinates scripts, narration and video composition. The Skill Evolution workflow tests revised skills against a validation set and accepts updates only when evaluation scores improve.

Similar to MS Agent