XiaomiMiMo/MiMo-Code

Features

• Terminal Coding Agents - Runs as a terminal coding agent that reads and edits project files, runs shell commands, and delegates work to LLM-powered agents.
• Terminal AI Coding Assistants - Reads and edits project files, runs shell commands, and manages version control from a single interactive terminal interface driven by AI agents.
• Terminal Coding Assistants - Reads and edits project files, runs shell commands, and manages Git from inside a terminal-based AI coding assistant.
• Agent Memory Persistence - Keeps project knowledge and decisions across sessions so the coding agent resumes with earlier context instead of starting over.
• Agent Session Memory - Stores project knowledge, session checkpoints, and task logs in a searchable database and injects them back when a session resumes.
• Context Memory Management - Controls what stays in the working context through automatic checkpoints and rebuilds so long sessions remain coherent within model limits.
• Context Window Management - Saves checkpoints and rebuilds context from memory plus recent messages when a model's window nears its limit, using per-model thresholds.
• Automatic Context Compaction - Saves checkpoints automatically and rebuilds working context when the model window nears its limit, using per-model thresholds.
• Multi-Agent Orchestration Frameworks - Spawns context-sharing subagents and runs scripted sandbox workflows that coordinate parallel agents with bounded retries and a judge model.
• Subagent Delegation - Delegates work to helper agents that share session context, run in parallel or background, and can be tracked and cancelled.
• Multi-Agent Task Orchestrators - Splits large jobs across parallel subagents and scripted sandbox workflows with bounded retries, with an independent judge confirming completion.
• Project Memory Retention - Keeps a persistent record of the codebase across sessions so the assistant recalls its understanding of the project later.
• Searchable Agent Memory Stores - Keeps project knowledge, checkpoints, notes, and task logs in a searchable database and injects them back into context when a session resumes.
• LLM Provider Configurations - Connects to major model providers via API keys, browser sign-in, or imported credentials, or any custom endpoint by base URL and model ID.
• Resumable Project Sessions - Injects stored project knowledge, checkpoints, notes, and task logs when a session resumes so work continues where it stopped.
• Database Memory Persistence - Stores project knowledge, checkpoints, notes, and task logs in a searchable database and reloads them when a session resumes.
• End-to-End Feature Delivery - Runs a contract-driven loop from clarification through specification, implementation, verification, and review, leaving feature documents in the workspace.
• Agent Lesson and Work Recording - Reviews past sessions to extract lasting knowledge into memory, prune outdated entries, and package repeated workflows into reusable skills.
• Agent Stop Guards - Prevents the agent from stopping until an independent judge confirms the stated session goal is met.
• Agent Task Orchestrators - Picks models and tools dynamically by task type and cost, and runs complex tasks across several agents working in parallel.
• Agentic Project Memories - Persists project knowledge, checkpoints, and task progress across sessions and turns past lessons into reusable skills.
• Autonomous Goal Pursuit - Repeats planning and acting cycles on its own until a stated goal is reached, with a judge model confirming completion.
• Build and Plan Agent Modes - Lets users switch with a key press between a full-permission build agent and a read-only plan agent for safe exploration.
• Agent Workflow Scripting - Runs multi-agent workflows as scripts in an isolated sandbox with fixed phases, bounded retries, and parallel branches for repeatable orchestration.
• Knowledge Distillation into Skills - Finds repeated manual workflows in recent sessions and packages the most confident ones into reusable skills, helper agents, or commands.
• Long Autonomous Runs - Keeps project knowledge, checkpoints, and task progress so long-running work resumes exactly where it stopped across sessions.
• Model API Provider Profiles - Registers any OpenAI-compatible provider by base URL, API key, and model ID, and declares which input types each model accepts.
• Multi-Provider Model Support - Works with many model providers from one terminal interface, with language server, tool server, and plugin support for extension.
• Persistent Task Trees - Tracks work as a numbered task tree whose progress is checkpointed so interrupted sessions resume at the exact node they left.
• Reusable Instruction Skills - Matches tasks to bundled or custom instruction sets by name and relevance, loading them automatically or through a slash command.
• Subagent Context Protection - Launches helper agents that inherit the parent's context, run in parallel or background, and are tracked in a registry supporting cancellation.
• Subagent Orchestrations - Launches helper subagents that share the parent's context, run in parallel or background, and report results back to the main agent.
• Project Knowledge Retention - Scans recent session traces, saves lasting knowledge to project memory, and removes outdated entries to keep context accurate.
• API-Key-Only Provider Connections - Guides first-launch setup to connect model providers through platform sign-in, imported credentials, API keys, or custom providers.
• AI Agent Permissions - Prompts before out-of-project actions, applies allow and deny rules, and offers optional auto-approval in disposable environments.
• Persisted Permission Rules - Passes every tool action through allow and deny rules, prompting for out-of-project access, with an optional auto-approve mode for disposable environments.
• Hierarchical Task Trees - Organizes work into a numbered task tree and saves progress at checkpoints so interrupted sessions resume at the exact node.
• Client-Server Architecture - Splits a lightweight local terminal client from a separate server that runs the session and accepts remote attachments by URL.
• AI-Driven Feature Implementation - Runs a contract-driven loop taking a feature from clarification through specification, implementation, verification, and review, leaving documents in the workspace.

Star history