Deckforge

What is Deckforge?

Deckforge is an API-first presentation generation server that converts structured JSON (Presentation Intermediate Representation) or natural-language prompts into professional PowerPoint (.pptx) or Google Slides decks. It is designed for both human users and AI agents, with built-in MCP tool integration for autonomous slide creation.

How to use Deckforge?

Deploy with Docker Compose, copy .env.example, run bash scripts/bootstrap-db.sh to initialize the database and obtain a test API key. Then call HTTP endpoints (/v1/render, /v1/generate, /v1/themes, /v1/slide-types, /v1/estimate) with the API key, or use the TypeScript SDK (@lukastan/deckforge) with a fluent builder pattern.

Key features of Deckforge

• 32 slide types and 24 chart types
• 15 built-in themes with custom brand kits
• AI content generation from natural language prompts
• MCP server with 6 tools for agent integration
• 5-pass QA pipeline with auto-fix engine
• TypeScript SDK with fluent builder

Use cases of Deckforge

• Generate executive board update decks from structured data.
• Create pitch decks from a single prompt for startups.
• Build finance-specific presentations with integrated charts.
• Automate recurring report generation for enterprise teams.
• Let AI agents produce themed, quality-checked slides autonomously.

FAQ from Deckforge

What slide types does Deckforge support?

32 slide types are supported, including 9 finance-specific slides (DCF summary, comp table, waterfall chart, deal overview, etc.) and 23 universal types (title, agenda, stats callout, chart, etc.).

Can Deckforge output to Google Slides?

Yes. Direct export via Google Slides API is supported (OAuth flow included), in addition to native PPTX output via python-pptx.

How do I get an API key for testing?

Run bash scripts/bootstrap-db.sh after setting up Docker Compose — it seeds a test user and outputs a test API key (dk_test_...) for local development.

Which LLMs can power the generation endpoint?

The /v1/generate endpoint supports Claude (Anthropic), OpenAI, Gemini, and Ollama — at least one LLM API key must be configured in .env.

Can I estimate the cost before rendering a deck?

Yes. The /v1/estimate endpoint computes credits and USD equivalent based on slide count, finance slide surcharges, and chart surcharges, so an agent can budget before calling /v1/render.