Choose theme
AI Resources
DeepCode
DeepCode is a coding agent from HKUDS with a public repository. Its terminal, desktop and browser interfaces share the same projects, sessions, models, skills, permissions, goals, and task history.
It can inspect and change a repository, run commands and tests, keep longer goals moving across interruptions, delegate focused work, and automate repeatable jobs. Its original Paper2Code workflow remains available for turning research papers and technical documents into working, testable projects. The v2.3.0 release connects all three to one local background service; closing an interface leaves its running work in that service. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.
What it is
A coding agent for real repositories
DeepCode reads and searches code, edits files, runs commands and tests, and keeps its work visible through plans, tool activity, diffs, results, and saved task evidence.
Why it stands out
The work can survive a restart or model change
Sessions preserve conversation, tool records, permissions, goals, and verification history. Longer goals can be paused, corrected, resumed, or continued through the same underlying workflow in the terminal, desktop or browser client.
Availability
Public source, CLI package, and desktop source path
The repository documents Python package installation for the CLI and source setup for the desktop app. Desktop release bundles are separate, so readers should check the current Releases page before expecting an installer for their platform.
Why it matters
What makes it useful
DeepCode keeps the work around a coding agent together: project rules, the conversation, tool calls, approvals, goals, tests, diffs, and completion evidence remain attached to the task. That gives developers a clearer way to resume a difficult change and see why the agent considers it finished.
What to know
Where it fits
DeepCode runs the agent and its task history, rather than only supplying a coding model. Its clients connect to a local service, so a developer can change interface while following the same work and answering its approvals. Paper2Code remains a separate research-reproduction path.
Notable points
What stands out
The DeepCode paper and repository report PaperBench results for scientific code reproduction. Those figures belong to the study's named subsets and setup; they are not a general performance ranking against continuously changing coding products.
Before using
What to review
The project documents permission modes for file edits and command execution. Its v2.3.0 release also lists model-egress controls, command screening, environment scrubbing, transparency logging and OS-keychain access as opt-in features, all off by default.
DeepCode connects to hosted providers, compatible endpoints or local models. Each chosen endpoint has its own credential, availability, cost and data-handling terms.
Project rules can live in ordinary repository files, while conversations and durable goals also depend on DeepCode's stored service state.
The desktop source instructions require platform toolchains; packaged installers and their platform coverage are listed separately on Releases.
Issue 128 reports unsafe command execution in earlier MCP tooling at commit b9ece60. That historical report does not establish the state of the current release; the v2 documentation separately describes its permissions and trust boundaries.
Start with the documented Read-only or Ask permission mode in a trusted or disposable repository; review the proposed diff and test results before accepting a change.
Decide which session history to retain, back up or remove. The upgrade recovery snapshot contains provider and MCP credentials, so keep it private; it excludes project working trees and cannot reverse file edits or shell commands. Restored Goals and interval Automations begin paused for explicit review.
Check the chosen provider's input handling before supplying sensitive code or using a shared machine; permission modes and opt-in controls do not settle that choice.
Reader fit
Who may find it relevant
Developers who want coding tasks, approvals, diffs, tests, and evidence to remain together across sessions.
People comparing one coding-agent workflow across a terminal and a desktop interface.
Researchers exploring paper-to-code reproduction with a public implementation and accompanying paper.
Teams experimenting with reusable skills, parallel agents, or recurring repository maintenance.
Less relevant for readers who want a no-setup consumer chatbot or a coding model checkpoint without an agent runtime.
Editorial note
Why LifeHubber lists it
A repository change can need investigation, test analysis and implementation review at the same time. DeepCode documents parallel agents working in isolated Git worktrees and returning their results to the main task for review and integration. That gives a developer a way to split those parts of a change while keeping the final merge decision together; conflicts still need review. Upgrading the agent involves more than preserving the repository being edited. DeepCode documents service prepare-upgrade as a way to stop its background service and snapshot its state before an upgrade. That makes session history and service state part of the upgrade task alongside the code files.
Source links
Source materials
Reader note
Before relying on this entry
LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.
What to explore next
Keep the project record useful outside one agent.
DeepCode can retain sessions, goals, rules, skills, and evidence. The next decision is what should also stay in ordinary project files and how another durable coding-agent workflow handles the same work.
More in Productivity
Keep browsing this category
Explore more productivity tools.
Penpot
penpot/penpot
An open-source collaborative design platform with an official MCP server that lets AI agents inspect and modify Penpot files, components, tokens, styles, layouts, and assets through hosted or local connection paths.
DeskcommCRM
melgarafael/DeskcommCRM
A self-hosted sales and support CRM centered on WhatsApp, with AI agents, lead pipelines, human handoff, follow-up rules, and a choice of AI providers. It uses a VPS and Supabase; AI features need a provider key.
Prime Agent
PrimeIntellect-ai/prime-agent
A Prime Intellect coding and research agent that uses persistent Python, child agents, saved working knowledge, and background sessions to keep larger tasks moving across terminal visits.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
