PM Agent Team

English | 简体中文

A multi-agent research automation platform for product managers

Competitive research · Interview analysis · PRD drafting · Social listening

Why it exists

Product managers spend a large amount of time on research workflows that are structurally repetitive but information-heavy. PM Agent Team turns those workflows into multi-agent pipelines that search, extract, analyze, write, review, and revise structured reports with citations.

Workflow	What the system produces
Competitive research	Market scan, product matrix, SWOT, user sentiment, and cited report
Interview analysis	Theme clustering, frequency analysis, source quotes, and confidence-ranked needs
PRD drafting	Background, goals, user stories, acceptance criteria, and risks
Social listening	Cross-platform posts, relevance filtering, clusters, and opportunity insights

What makes it more than a single LLM call

• Parallel agent pipeline — planning, search, crawling, social listening, extraction, analysis, writing, and review.
• Self-correction loop — low review scores send critique back to the writer for one revision pass.
• Eleven expert roles — research, market intelligence, lean validation, risk review, and other specialists are selected by scenario.
• Project memory — related tasks share a project space and can recall prior research.
• Incremental follow-ups — a follow-up becomes a child task and merges back into the parent report.
• Cost observability — token and cost attribution by task, agent, and model with an optional budget cap.
• Failure recovery — exponential LLM retry and non-blocking optional agents.
• Slack and Jira integration — send report cards, create issues, or trigger PRD drafting from Jira webhooks.

Interface

The React interface includes a workflow dashboard, scenario pages, task lists, a live agent timeline, report previews, review scores, follow-up controls, and social-post exploration.

Architecture

flowchart LR
    U[Product request] --> P[Planner]
    P --> R{Parallel research}
    R --> S1[Search]
    R --> S2[Scraper]
    R --> S3[Social listening]
    S1 & S2 & S3 --> E[Extractor]
    E --> A[Analyzer]
    A --> W[Writer]
    W --> RV[Reviewer]
    RV -->|score >= 7| OUT[Report]
    RV -->|score < 7| W2[Rewrite with critique]
    W2 --> OUT

Loading

The backend is a Go HTTP/SSE service with an in-memory worker queue and store. The frontend is React 18, Vite, TypeScript, React Router, and Tailwind CSS. PostgreSQL/pgvector migrations are included as a future production path, while the default development setup works without them.

Quick start

1. Clone and configure

git clone https://github.com/realruian/PM_Agent_Team.git
cd PM_Agent_Team
cp .env.example .env

Add a supported LLM provider key to .env. With no key, the application falls back to mock mode so the UI and pipeline can still be explored.

2. Start the backend

Requires Go 1.26.2.

set -a && source .env && set +a
cd server
go run ./cmd/server

The API listens on http://localhost:8080 by default.

3. Start the frontend

cd web
npm install
npm run dev

Open http://localhost:5173.

Main scenarios

• requirement_analysis
• competitor_research
• requirement_validation
• interview_analysis
• prd_drafting
• social_listening

Tasks stream agent activity over SSE and expose the final report, sources, review metadata, social posts, traces, and usage data through the API.

Configuration

See .env.example for the complete list.

Variable	Purpose
LLM_PROVIDER, LLM_MODEL, LLM_BASE_URL	Model provider and optional custom gateway
Provider API keys	Configure one provider; no key enables mock mode
MOCK_MODE	auto, always, or never
TASK_BUDGET_USD	Per-task model budget
SEARCH_PROVIDER	Search routing; Tavily and Jina keys are optional
SLACK_WEBHOOK_URL	Slack notifications
Jira variables	Jira REST and webhook integration
Social credentials	Optional X, Douyin, TikTok, or YouTube access
HTTP_ADDR, CORS_ORIGINS	HTTP listener and allowed frontend origins

Repository layout

server/              Go API, agents, LLM adapters, tools, store, jobs, and SSE
web/                 React dashboard and workflow pages
assets/              Logo, banner, screenshots, and interactive architecture
docs/                Data-pipeline design notes
docker-compose.yml   PostgreSQL and pgvector production path
.env.example         Configuration template

Verification

cd server && go test ./... -race
cd web && npm install && npm run build

Roadmap

Planned work includes pgvector-backed memory and evaluation, multimodal input, team review workflows, scheduled industry data ingestion, and a public beta.

License

MIT · PM Agent Team v0.6