- Category
- AI Agents
- Rank
- No. 229Tools index
- Pricing
- Open Source
- Type
- AGENT
- Builder
- hkuds
- GitHub
- 16.5k stars
- Latest release
- v2.2.0
- Date
About
Open agentic coding system from HKUDS — turns research papers, text descriptions, and design intent into working web frontends and backend services.
What it does
DeepCode is a local coding-agent runtime with shared terminal and desktop interfaces. It keeps projects, session history, goals, permissions, skills, models, automations, code changes, and verification in one workspace. The runtime selects an LLM provider, exposes controlled coding tools, executes iterative tasks, and records durable results.
Why it's ranked here
DeepCode merits attention because its architecture treats agent operation as a governed workflow, not a single prompt. Shared runtime wiring keeps terminal, headless, and desktop behavior aligned. Permission checks, bounded verification, private state storage, provider selection, MCP support, and persistent sessions make the engineering unusually concrete. The breadth also raises setup and operational complexity.
What's good
Security controls reach the execution path: tool use passes through permission decisions, interactive approvals fail closed, and code orchestration does not bypass policy. Verification discovers Python, Node, and Rust tests through an allowlist, limits captured output, applies timeouts, and terminates process groups. Provider and project configuration can be layered while keeping user credentials outside project settings.
Tradeoffs
The core install requires Python 3.12 or newer and pulls a broad dependency set spanning several model SDKs, MCP, document handling, terminal UI, networking, and reporting. Desktop code is excluded from the Python source distribution, so the package install does not deliver every interface described. Effective use also requires choosing and configuring a model provider or local model service.
How to use it well
Use DeepCode when you want a coding agent to work repeatedly inside local projects with recorded sessions, explicit permissions, reusable skills, MCP tools, and built-in test execution. It fits developers who value inspectable tool activity and shared behavior across terminal, headless, and desktop workflows. It does not replace model access, provider credentials, or dedicated desktop distribution.
Technical notes+
setup.py builds the deepcode-hku package with setuptools, reads dependencies from requirements.txt, requires Python 3.12+, and installs the deepcode and deepcode-app-server console entry points. core/config.py defines layered user and project JSON configuration for providers, connections, MCP servers, skills, security, workspace limits, and logging. core/agent_setup.py assembles the shared AgentSession, native tools, permission engine, hooks, code mode, optional delegation, and frontend callbacks. core/verification.py performs bounded discovery and execution for pytest, unittest, npm, and Cargo tests. core/private_storage.py creates private state directories and regular files with symlink and identity checks. pyproject.toml configures pytest to use tests while excluding tests, desktop, docs, and development material from the Python source archive.
Observed
- License
- MIT
- Language requirement
- Python 3.12 or newer
- Packaging
- Setuptools package named deepcode-hku with dependencies read from requirements.txt
- Command interfaces
- Console entry points provide deepcode and deepcode-app-server
- User interfaces
- Interactive CLI and Tauri Desktop workbench share one Agent runtime
- Extension interface
- MCP supports stdio, SSE, and Streamable HTTP server configurations
- Platform support
- Package metadata declares operating-system independence, with explicit POSIX and Windows runtime handling
- Test structure
- Pytest is configured for a tests directory, which is excluded from the Python source archive
Read from README.md, setup.py, pyproject.toml, requirements.txt, __init__.py, core/config.py, core/version.py, core/__init__.py, core/file_lock.py, core/reasoning.py, core/agent_setup.py, core/llm_runtime.py, core/verification.py, core/platform_compat.py, core/private_storage.py.
What it can do
Convert research papers into working code
Research paper documents → Functional web application code
Generate web frontend from text descriptions
Natural language description of desired interface → Complete frontend code and components
Create backend services from design specifications
Design intent and system requirements → Backend service code and APIs
Transform academic concepts into executable software
Theoretical concepts from papers → Working software implementation
Generate full-stack applications autonomously
High-level project requirements → Complete web application with frontend and backend
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.