Skip to content

architecture.md

Latest commit

 

History

History
50 lines (41 loc) · 2.63 KB

architecture.md

File metadata and controls

50 lines (41 loc) · 2.63 KB
Raw

Architecture: Code Editors & IDE Assistants

How the code editors & ide assistants landscape is shaped, and how the Quest tests it.

The landscape at a glance

Tool Tier License Focus Notes
Cursor A Proprietary AI-native IDE VS Code fork; market leader; $20-200/mo
GitHub Copilot A Proprietary IDE assistant Widest adoption; usage-based credits from June 2026
Claude Code A Proprietary CLI agent ~$2.5B ARR; 1M context; Agent Teams; SWE-bench 82%
Windsurf A Proprietary AI-native IDE Formerly Codeium; acquired by Cognition Dec 2025
OpenCode A MIT Terminal TUI 150K+ stars; 75+ LLM providers; Go-based
Cline A Apache-2.0 VS Code + CLI 5M+ installs; native subagents; BYOK; browser automation
Aider A MIT Terminal pair-programmer 42K+ stars; Git-aware diffs; model-agnostic
Continue A Apache-2.0 IDE extension 20K+ stars; BYOK; code review; CI-enforceable checks
Zed A Open source Rust-native editor Multi-provider chat; real-time collab; GPU-accelerated
Trae A Proprietary AI IDE ByteDance; free; VS Code fork; 95% WeChat SDK accuracy
Tabnine A Proprietary Completion + chat 1M+ devs; on-prem; zero data retention; SOC 2/GDPR
Codex CLI A Apache-2.0 CLI agent OpenAI; Rust-native; 240+ tok/s; included in ChatGPT
Gemini CLI A Apache-2.0 CLI agent 1M token context; 1,000 req/day free; full open source
Kimi CLI A Open source Terminal agent Moonshot; Agent Swarm up to 300 sub-agents; K2.6
Devin A Proprietary Autonomous SWE Cognition; full VM; Slack-native; $20/mo entry + ACUs
OpenHands A MIT Autonomous agent 70K+ stars; formerly OpenDevin; Docker sandboxed
CodeRabbit A Proprietary AI code review $550M valuation; 8,000+ customers; line-by-line feedback

How the Quest tests a tool

Same harness for all entries; the judge was frozen before any tool ran:

Adapter[frozen CategoryAdapter contract]
  ├── setup()    → install, configure
  ├── load()     → ingest workload
  ├── await_ready() → async barrier
  ├── query()    → run test, get response
  └── teardown() → cleanup, measure
       ↓
Telemetry: latency · tokens · $ · ops notes
       ↓
Grading: deterministic + LLM judge (frozen prompts)
       ↓
Raw results JSON (published)

The await_ready() barrier is where async designs get their cost measured instead of hidden.

License

Content is licensed CC BY 4.0 — share and adapt with attribution to ArdurAI / Code Editors & IDE Assistants Almanac.