One category, two kinds. Coding agents (27) write code inside your IDE or terminal; agent platforms (15) are for building, orchestrating and shipping agents. Switch with the tags below.
Showing 27/27
Agent Index
AGENT INDEX27 ENTRIES
No.001AI-native IDEClosed
Cursor
Anysphere
An AI-native IDE deeply customized from VS Code. Its Agent mode supports cross-file refactoring, test verification and autonomous task execution; launched a proprietary coding model in 2026.
Anthropic's terminal-side AI coding agent with top-tier reasoning. Supports Multi-Agent parallelism, MCP toolchains and Hooks automation; scores 80.8% on SWE-bench.
Large codebase refactoringComplex multi-step reasoningAutonomous bug-fix + test loop+1
No.003IDE PluginClosed
GitHub Copilot
Microsoft / GitHub
An IDE-plugin AI assistant jointly built by Microsoft and GitHub. Fully agentified in 2026 and shifted to usage-based billing, deeply integrated with the GitHub ecosystem (Issue/PR/repo indexing).
Agent modeCode completionPR generation
Context window
Repo-level indexing
Models
4
Free tier
✓ Yes
Billing
Usage-based (June 2026 reform)
Pro$10/mo (switching to usage-based from June 2026)
Google's free terminal AI coding agent with 1000 free requests per day, a 1M context window and MCP support. At I/O 2026, launched Antigravity 2.0 to rival Claude Code.
The world's first autonomous AI software engineer, independently completing the full flow from requirement understanding to PR submission. Scores over 50% on SWE-bench Pro, valued at $26 billion.
Agent modeFully autonomousSelf-learningMulti-repo
Context window
Multi-repo level
Models
3
Free tier
✓ Yes
Billing
Free + subscription (allowance refreshes daily/weekly; overage billed at API rates)
Pro$20/mo (higher allowance, OpenAI / Claude / Gemini frontier models and cloud agent; extra usage at API rates)
End-to-end bug fixingCodebase migrationMulti-week large projects+1
No.007Desktop AgentNEWOpen
DeepSeek Harness
DeepSeek
DeepSeek's open-source coding agent framework released on August 13, 2026, built on the Cordis plugin meta-framework (TypeScript, MIT). It passed 70K GitHub Stars within a month and reached 246K by October — the fastest-growing open-source project of the year. v0.2 preview added desktop builds (Sept 29), the official desktop release covers macOS and Windows on Oct 2 (Linux via the @deepseek-ai/dsh npm package), and an Oct 3 update strengthened plugins and cross-platform support plus an experimental Claude Code Mods compatibility layer. Ships a web UI and Python SDK; still developer preview with breaking changes expected.
Auditable daily codingExtending it via pluginsLocal/self-hosted models+1
No.008Terminal AgentOpen
opencode
Anomaly
An open-source terminal AI coding agent from Anomaly, available as terminal, IDE and desktop clients. Supports multi-session workflows, any model provider (including local models), and can reuse a GitHub Copilot / ChatGPT subscription or your own API key. Vendor-neutral and self-hostable.
Agent modeMulti-file editTerminal execution
Context window
—
Models
1
Free tier
✓ Yes
Billing
Free open source + your own model costs
Code PlanOpen · free
Auditable daily codingSwitching freely between modelsLocal model integration+1
No.009IDE PluginOpen · Apache 2.0
Cline
Open-source community
An open-source VS Code coding agent plugin. Supports autonomous file operations, terminal command execution and browser automation; works with any LLM, backed by a rich community ecosystem.
Autonomous coding inside VS CodeBrowser automation testingMCP toolchain extension+1
No.010AI-native IDEClosed
TRAE
ByteDance（ByteDance）
ByteDance's AI-native IDE, the first of its kind in China. The basic edition is permanently free with 98% code-generation accuracy, supporting SOLO/Builder/IDE triple modes; over 6 million registered users.
Agent modeImage to codeChina-optimizedSOLO modeBuilder mode
Context window
Whole-project (CUE engine)
Models
4
Free tier
✓ Yes
Billing
Free + token subscription (adjusted Feb 2026)
Pro$19/mo (approx. $10/mo during promo)
◈ Free tier: Basic edition permanently free (no AI call limit)
An MIT-licensed open-source AI software engineering agent platform (formerly OpenDevin). Offers a web Agent Canvas, OpenHands Cloud and full self-hosting (local, private cloud, VM, enterprise VPC). Runs in isolated sandboxes that edit files, execute commands, run tests and open PRs, and can work alongside Claude Code, Codex or Gemini CLI.
Self-hosted AI developerSandboxed autonomous edits and PRsEnterprise code staying in-network+1
No.012Terminal AgentOpen
Qwen Code
Alibaba (QwenLM)
Alibaba Qwen team's open-source terminal coding agent (Apache-2.0): native CLI plus VS Code / Zed / JetBrains extensions and an experimental web UI. Switch providers at runtime between Qwen API, OpenAI, Anthropic, Gemini or local models, with Telegram / DingTalk / Feishu channels, headless mode and CI support.
Coding with the Qwen ecosystemCompliance-sensitive developmentPrivate local-model setups+1
No.013Terminal AgentOpen · Apache 2.0
Aider
Open-source community（Paul Gauthier）
An open-source terminal AI coding agent with native Git integration. Supports multi-model switching and auto-commits changes; an active community, ideal for developers who prefer the command line.
AWS's Agentic AI IDE, pioneering Spec-driven development (requirements → design → tasks → code). Multi-model Auto agents automatically match the best execution strategy; deeply integrated with the AWS cloud ecosystem, delivering one-stop from prototype to production-grade code.
Agent modeSpec-drivenAWS integrationRequirements to codeMulti-model orchestrationAuto agent
Google DeepMind's Agent-first development platform, first launched in November 2025 and upgraded to 2.0 at I/O 2026. Natively integrates Gemini 3 Pro, supports multi-agent parallel orchestration, and autonomously operates the terminal and browser — fully rivaling Claude Code and Codex.
An AI coding assistance platform built for large codebases, with a 200K-token ultra-long context window. Deeply indexes team private codebases and commit history, prioritizing existing in-house utilities to avoid reinventing the wheel. Raised a $227M Series B in 2025 at a $977M valuation.
Agent modeAuto refactoringUnit test generationTeam codebase indexingMulti-file analysisIP protection+1
Context window
200K tokens (incl. full commit history)
Models
2
Free tier
✓ Yes
Billing
Free + subscription
Pro$19/mo
◈ Free tier: Free tier available (limited quota)
Large legacy codebase understandingTeam knowledge reuse (avoid reinventing the wheel)Cross-file automated refactoring+2
No.017Terminal AgentClosed
JetBrains Junie
JetBrains
JetBrains' official AI coding agent. It runs "in the IDEs" inside IntelliJ IDEA, PyCharm, WebStorm and peers, and ships a standalone Junie CLI for terminal and GitHub/GitLab workflows. Deeply wired into JetBrains' indexing and refactoring engines, with BYOK support for external APIs and local models, metered in AI Credits.
ProAI Pro $8.33/user/mo (10 AI Credits per 30 days, top-ups available)
◈ Free tier: Junie Lite free (no credit card, includes free models)
Java/Kotlin projectsJetBrains IDE integrationIn-IDE refactoring+1
No.018AI-native IDEClosed
CodeBuddy
Tencent（Tencent Cloud）
Tencent Cloud's intelligent coding assistant for the AI era, achieving full coverage across IDE, plugin and CLI (CodeBuddy Code). Its Craft Agent mode lets AI autonomously understand requirements and rewrite multiple files; the first MCP-supporting code assistant in China, integrating Hunyuan + DeepSeek dual models.
Generate a full project from one sentenceAutonomous multi-file rewritingDesign-to-code+2
No.019AI-native IDEClosed
Qoder (ex-Tongyi Lingma)
Alibaba Cloud（Alibaba Cloud）
Alibaba Cloud's intelligent coding assistant, renamed and upgraded in May 2026 into an autonomous agent development workbench. Supports agent-team autopilot and enterprise-grade security compliance.
Moonshot AI's multimodal AI coding agent, built on the K2.7 Code model (1T params / 32B active MoE). Supports image/video input-driven coding; reaches 81.1% on MCP Mark Verified, surpassing Opus 4.8. Open-sourced model weights in June 2026.
Agent modeImage to codeVideo to codeMultimodal input
Xiaomi's MiMo team open-sourced terminal AI coding agent, built on OpenCode under the MIT License. Its pioneering four-layer persistent memory system (Session/Project/Global/History) solves long-session amnesia; Dynamic Workflow can schedule hundreds of parallel sub-agents. Gained 11K GitHub stars in 16 days.
Long-horizon tasks (hundreds of continuous steps)Cross-session memory (understands your project better over time)Dynamic multi-agent parallelism+2
No.022AI-native IDEClosed
Factory AI
Factory
An agentic software development platform built around Droid, covering coding, testing, review, deployment and enterprise engineering automation. Offers desktop, CLI, SDK and cloud background agents that can run long asynchronous tasks for engineering teams.
Enterprise SDLC automationDispatching long async tasksCode review and deployment+1
No.023Desktop AgentClosed
Zencoder
Zencoder
An enterprise codebase-aware AI coding agent platform with the web-based Zenflow, a desktop app and IDE extensions. Deep codebase understanding, multi-repository intelligence and 2,000+ integrations (GitHub, Slack, Google Workspace, Microsoft 365, Notion, HubSpot), with on-premise and hybrid deployment options.
◈ Free tier: 7-day trial (5,000 credits, Pro features included)
Large codebase understandingMulti-repo developmentPrivate enterprise deployment+1
No.024Terminal AgentClosed
Amazon Q Developer
Amazon Web Services（AWS）
AWS's AI coding assistant, deeply integrated with the AWS ecosystem. Supports code transformation (Java 8→17), security scanning and enterprise compliance; fully agentified in 2026.
AWS cloud-native developmentJava version upgrade migrationEnterprise security compliance scanning+1
No.025Terminal AgentClosed
Amp
AmpCode (Ampl Labs)
A terminal AI coding agent from the team behind Sourcegraph, now operating independently as AmpCode / Ampl Labs. Built on strong code search and codebase context, focused on an agentic coding workflow in the editor and terminal, with Hobby free tier plus Megawatt and Gigawatt plans.
Agent modeMulti-file editTerminal execution
Context window
—
Models
0
Free tier
✓ Yes
Billing
Subscription
ProMegawatt $20/mo
◈ Free tier: Hobby free ($0)
Coding with large codebasesCross-file refactoringTerminal-first agent workflows+1
No.026AI-native IDEClosed
Devin Desktop (ex-Windsurf)
Cognition AI（formerly Codeium）
The former Windsurf IDE, renamed after acquisition by Cognition. Blends the Cascade agent engine with Devin capabilities, supporting hybrid local + cloud agent management.
SpaceXAI's (formerly xAI) terminal AI coding agent CLI, in early beta since May 2026 — the company's first Claude Code / Codex competitor. It offers plan mode, up to 8 parallel sub-agents, MCP tool extensions and human approval steps, and works with a SuperGrok subscription or the xAI API. Grok 4.5 / 4.6 meanwhile push long-running agent and coding capabilities.