Open Source Model

Aider

✓

Open-source terminal pair programmer with git-native workflows

Aider is an open-source AI pair programming tool that runs in the terminal and keeps your workflow inside git. It can automatically create commits, apply diff-based edits across multiple files, and map your repository for large projects. Aider is model-flexible: you can use current OpenAI, Anthropic, Google, DeepSeek, and local model stacks through direct APIs, OpenRouter, Ollama, or compatible endpoints.

• ●Automatic git commits with descriptive messages
• ●Diff-based, multi-file editing directly in the terminal
• ●Works with current frontier model families (OpenAI, Anthropic, Gemini) plus DeepSeek and local models

AutoGen

✓

Open-source multi-agent framework (Python/.NET) with AgentChat and Studio

AutoGen is Microsoft's open-source framework for building agentic applications with your own model API keys. The current stable release (python v0.7.5, Sep 30 2025) ships a layered architecture: Core for event-driven message passing across Python and .NET runtimes, AgentChat for common multi-agent patterns, Extensions for model/tool connectors (OpenAI and Azure OpenAI, Anthropic, Gemini via OpenAI-compatible endpoints, and local models via Ollama), plus AutoGen Studio for no-code multi-agent prototyping and AutoGen Bench for evaluation. It's MIT-licensed and maintained in the open on GitHub.

• ●Layered Core, AgentChat, and Extensions architecture with event-driven agents
• ●AutoGen Studio for visual multi-agent workflow building
• ●AutoGen Bench and Magentic-One reference agents for evaluation and examples

Cline

Open-source autonomous AI coding agent for VS Code with plan-and-act workflows

Cline is an open-source AI coding agent that runs as a VS Code extension. Originally released as Claude Dev, it reads codebases, creates and edits files, runs terminal commands, and can automate browser interactions with user approval at each step. Cline supports plan mode for analysis without changes and act mode for executing edits. It works with any LLM provider including Anthropic Claude, OpenAI, Google Gemini, AWS Bedrock, Azure OpenAI, and local models via Ollama or LM Studio, with deep MCP support for custom tools.

• ●Plan mode: analyze requests and propose approaches without modifying files
• ●Act mode: execute changes with approval to prevent unintended rewrites
• ●Reads codebase, edits files, runs terminal commands, automates browser

Continue

✓

Open-source, model-agnostic AI coding assistant for VS Code and JetBrains

Continue is an open-source coding assistant that lets developers use their own models (OpenAI, Anthropic, local via Ollama/LM Studio) with custom prompts and workflows. It runs as a VS Code and JetBrains extension, supports chat and completions, and can be self-hosted or run fully local for privacy.

• ●Model-agnostic with BYO keys
• ●Open-source and customizable prompts/commands
• ●Local/offline support via Ollama or LM Studio

CrewAI

✓

MIT-licensed agents, crews, and flows framework with optional control plane

CrewAI is an MIT-licensed Python framework for multi-agent automations. The 1.6.1 release (Nov 29 2025) ships Crews for autonomous collaboration and Flows for event-driven orchestration with guardrails, memory, knowledge, and observability hooks. It runs with your own model/API keys (OpenAI/Azure OpenAI, Anthropic, OpenAI-compatible providers like Groq, or local via Ollama/LM Studio) and offers an optional Crew Control Plane for hosted/on-prem tracing, triggers, and RBAC.

• ●Crews for autonomous agent collaboration and Flows for event-driven orchestration
• ●Guardrails, memory, knowledge, and structured outputs for production automations
• ●CLI/SDK with uv-based project scaffolding and tool integrations

DeepSeek Coder

✓

Open-source MoE coding model (V2) with 128K context

DeepSeek Coder V2 is an open Mixture-of-Experts coding model (16B Lite and 236B) with a 128K context window and support for 338 programming languages. Weights are published on Hugging Face for self-hosting; code is MIT-licensed and the model weights use the DeepSeek Model License. Hosted inference is available through the DeepSeek API using OpenAI-compatible endpoints.

• ●DeepSeek-Coder-V2 Mixture-of-Experts models (236B total / 21B active, plus 16B Lite / 2.4B active)
• ●128K context window with fill-in-the-middle support
• ●Supports 338 programming languages

Goose

Open-source terminal-first AI agent from Block for autonomous coding

Goose is an open-source AI agent from Block that works as a terminal-first coding assistant. Rather than autocomplete, it executes tasks autonomously in your terminal: running commands, editing files, making multi-file changes, reading error logs, and self-correcting when commands fail. Goose supports multiple providers including OpenAI, Anthropic, Google, Meta, and Ollama, and integrates with Claude Code or Google Gemini CLI subscriptions. It includes MCP support for tools, APIs, databases, and browser automation, with session management for conversation context and state.

• ●Terminal-first: executes commands, edits files, runs tests
• ●Self-corrects when commands fail
• ●Multi-file editing and git commands

LangChain

✓

Open-source framework for LLM apps with LangSmith observability

LangChain is an open-source framework for building LLM apps with composable chains, agents, tools, and RAG. It supports Python and TypeScript, integrates with major model providers and vector stores, and pairs with LangSmith for tracing, monitoring, and evals.

• ●Composable chains for multi-step workflows
• ●Agent framework with tool use
• ●RAG (Retrieval Augmented Generation)

LangGraph

✓

Graph-based framework for controllable, stateful AI agents

LangGraph is an open-source framework in the LangChain ecosystem for building stateful, multi-step agents with controllable graphs (single, multi-agent, hierarchical flows). It supports human-in-the-loop interrupts, streaming, memory, and tool use, and can be deployed via LangGraph Platform or self-hosted OSS.

• ●Graph-based agent architecture (single, multi-agent, hierarchical)
• ●Stateful execution with memory and persistence
• ●Branching and conditional control flows

Ollama

✓

Run AI models locally with Docker-like simplicity, 200+ model families, and full API compatibility

Ollama is the open-source standard for running AI models locally. With 155K+ GitHub stars and 2.5M weekly downloads, it provides Docker-like commands (pull, run, create) to manage 200+ model families on your own hardware. Ollama offers full OpenAI and Anthropic API compatibility as a drop-in replacement, GPU acceleration across Apple Metal, NVIDIA CUDA, AMD ROCm, and Vulkan, plus features like vision support, structured JSON outputs, tool calling, embeddings, and experimental image generation. Ollama Turbo ($20/month) adds optional cloud inference for users who want both local and hosted options.

• ●200+ model families: Qwen2.5-Coder, DeepSeek-Coder V2, Codestral, Qwen3-Coder, GPT-OSS, Llama 4, and more
• ●Docker-like CLI: ollama pull, ollama run, ollama create with Modelfile customization
• ●OpenAI and Anthropic API compatibility as a drop-in endpoint replacement

OpenCode

Open-source AI coding agent for the terminal with multi-provider support

OpenCode is an open-source AI coding agent built for the terminal. It helps developers write code via a terminal interface, desktop app, or IDE extension. OpenCode supports many LLM providers including Claude, GPT, Gemini, and local models, with GitHub Copilot and ChatGPT Plus login options. It features LSP integration, multi-session agents, and a privacy-first approach that does not store code or context. The CLI supports run commands for single tasks, a headless API server, and MCP management.

• ●Terminal interface and desktop app (beta)
• ●opencode run for single-task execution
• ●opencode serve for headless API

Roo Code

Open-source AI-powered autonomous coding agent for VS Code

Roo Code, formerly Roo Cline, is an open-source AI-powered autonomous coding agent that integrates into VS Code. It acts like a virtual dev team in your editor, with natural language communication, multi-file editing via diff-based changes, terminal command execution, and automated browser actions. Roo Code offers specialized modes for coding, debugging, and architecture, and is model-agnostic with support for any OpenAI-compatible or custom API. It requires user approval before executing automation. The extension includes a marketplace for community MCPs and Modes.

• ●Natural language communication with AI agent
• ●Multi-file editing with diff-based changes
• ●Terminal command execution and browser automation

StarCoder 2

✓

State-of-the-art open-access code LLM by BigCode

StarCoder 2 is an open-access code LLM family (3B/7B/15B) from BigCode (Hugging Face & ServiceNow), trained on The Stack v2. Weights are released under OpenRAIL-M for local or hosted inference.

• ●Open-access weights (OpenRAIL-M)
• ●3 sizes: 3B, 7B, 15B parameters
• ●Trained on The Stack v2

Zed

High-performance Rust code editor with agentic AI and open-source edit prediction

Zed is a high-performance code editor built in Rust with integrated AI capabilities. It runs at 120fps with minimal latency and uses multiple CPU cores and GPU efficiently. Zed offers agentic editing where AI agents can search codebases, make changes, and write code with real-time review. It includes Zeta, an open-source language model for edit prediction, inline assistant for sending code to LLMs, and text-based chat. Zed supports Claude, OpenAI, Anthropic, Ollama, and other providers. It features native Git support, multiplayer collaboration, and is open-source under GPL v3.

• ●Agentic editing: AI agents search, edit, and write with real-time review
• ●Zeta: open-source edit prediction model
• ●Inline assistant: send code to LLMs for transformation