Follow the agent team
A LangGraph workflow coordinates ten roles. The application visualizes real execution events, agent inputs and outputs, revisions, and human review.
Static project showcase. No AI generation, backend, accounts, or saved campaigns run on this page. Run the full application locally to try the workflow.
Built by Nikhil Jadhav
Good content.
Clear evidence.
A human final say.
Meet ContentForge AI: a multi-agent workspace that connects research, writing, evidence review, and approval in one inspectable content pipeline.
Open project · No-key local demo available
From brief to approved contentArchitecture illustration
Your content briefAudience · tone · platforms · sources
ResearchStored source passages
StrategyAngle and outline
Draft → inspect → reviseWriting, evidence, brand, channels, and quality
Human review & approvalApprove the exact versions before export
Conceptual flow, not a running campaign or live agent status.
Inside the full application
Every draft has a review trail.
Explore these features in the repository or local application. This static page explains the project without connecting to a server.
Inspect the claims
Trace claims to stored passages, inspect evidence assessments, and review warnings. Evidence matching is a transparent heuristic, not independent fact certification.
Approve what gets exported
Edit versioned drafts with formatting, review each version’s evidence, and export approved content as Markdown, JSON, or a ZIP package.
Blog articlesLinkedIn postsInstagram captionsYouTube scripts
Ten specialist roles
Open a role to see its responsibility.
These descriptions explain the implementation. Opening a card does not start an agent or make an AI request.
01 · Supervisor
Turns the brief and constraints into a bounded execution plan. The graph owns routing; model outputs cannot invoke arbitrary tools.
02 · Research
Reads stored source passages and identifies relevant findings, source references, and gaps. Autonomous web search is outside the current implementation.
03 · Strategy
Shapes the campaign’s angle, outline, audience, and tone from the brief and its constraints.
04 · Writer
Creates the requested platform drafts using the strategy and available evidence, and responds to revision feedback.
05 · Fact Checker
Extracts claims and conservatively matches them to passages. Missing or conflicting evidence stays visible for review.
06 · Editor
Refines complete drafts and records editorial notes while preserving version history.
07 · Brand Guardian
Checks drafts against the brand profile and adjusts their voice and presentation.
08 · SEO & Channels
Adapts drafts and metadata for the selected content platforms. It does not publish to social networks.
09 · Quality
Uses an explainable rubric and bounded revision routing. A high score does not remove factual warnings.
10 · Export
Packages the exact versions approved by a person, with evidence references and checksums. Calendar entries are reminders, not automatic publication.
Try the working software
Run the no-key demo on your computer.
Install Git and Docker with Linux containers, then use these commands. Open http://localhost:7860 and choose Explore the live demo.
The demo runs real orchestration with labeled sample outputs. Live AI and email are disabled. This temporary container loses its saved data when removed—download your work first.
Read the full setup guidegit clone https://github.com/nikhiljadhav6869-sudo/ContentForge-AI.git
cd ContentForge-AI
docker build -t contentforge-demo .
docker run --rm -p 127.0.0.1:7860:7860 contentforge-demoNext.jsFastAPILangGraphPostgreSQL / SQLiteRedis / local executorDocker