Skip to content

Roadmap

What's done and what's coming.

Done

  • 5KB kernel + 16 composable tools
  • 7 channels (CLI / HTTP / Feishu / WeCom / Telegram / GitHub / Webhook)
  • A2A protocol (agent-to-agent)
  • GEPA self-evolution
  • 200-task benchmark suite (AgentBench + GAIA) at 89.0% pass rate
  • Bilingual README (English + Simplified Chinese)
  • Interactive HTML benchmark report
  • Issue / Discussion templates

In Progress

  • Standalone docs site (MkDocs Material on GitHub Pages)
  • npx quark-agent one-line installer
  • Animated demo GIF for README hero

Planned

  • Plugin marketplace (quark-agent add @someone/slack-skill)
  • Multimodal computer-use tool (Claude Computer Use style)
  • Discord channel
  • Slack channel
  • Streaming function-call benchmark (τ-bench style)
  • SWE-bench subset for code engineering tasks
  • Local model adapters (Ollama, llama.cpp) with first-class support
  • Web UI v2 with task timeline visualization

Long-term

  • Browser-use agent (Playwright-backed)
  • Long-running background agents with persistence
  • Federated A2A network (registry + discovery)
  • Self-hosted eval dashboard

How to Influence

Open a Discussion tagged ideas — small, concrete proposals get traction fast.