AGENTICHUB

Autonomous development platform — from GitHub issue to production-ready Pull Request

ISSUE → PR PIPELINE

1 CLICK

Delegate a task to the agent — the system explores the repo, writes code, and opens a PR.

TWO AGENTS

PLANNER + ENG

Planner writes the plan without shell access. Engineer implements code, tests, and gh pr create.

STACK

LangGraphLangChainNext.jsDockerPostgreSQLQDrant

WHAT WE BUILT & HOW IT CONNECTS

Agentic Hub is a production-grade NestJS API with an in-process LangGraph runtime. Agents run inside the same Node.js process as the REST API — no separate Python worker.

LangGraph issue-execute: StateGraph with classify → plan → validate → execute → extract PR URL. One-shot for simple tasks, full-plan for complex ones.
Daytona sandbox: isolated VM per project. The agent works only on feature branch agent/task-{id}; push to default branch is blocked.
QDrant: semantic code index + episodic memory — the agent does not start every task from zero.

HOW IT WAS IMPLEMENTED

PHASE 1: WEBHOOK → DELEGATE → START

Issue in GitHub/GitLab → webhook creates a Task. Delegate marks the issue in VCS. BullMQ enqueues task-start.

PHASE 2: PLAN → VALIDATE → EXECUTE → PR

Planner explores the repo and writes the plan. Validator checks structure. Engineer implements by phases. PR URL is extracted and stored in PostgreSQL with session logs.

AGENTIC HUB CONSOLE

Enter a task description to simulate the pipeline (for example, "fix the typo in README" or "add an export endpoint"):

[ PIPELINE STEP LOGS ]:

Waiting for a task description to start the pipeline.