Self-Hosted Design Tools
As of August 2026, AIDiveForge tracks 8 self-hosted design tools. The top three by verified-data score are Drever, GenUI, and Framesmith. Curated self-hosted design tools tracked by AIDiveForge. Listings are verified against each tool's live website and re-checked regularly.
Last updated July 28, 2026 · 8 tools
Ranked by AIDiveForge's verified-data score: data completeness, verification recency, community rating, and real visitor engagement. How we rank · No tool can pay for placement.
1. Drever
The core workflow is brief-in, plan-review, then slide authoring — the agent pauses after generating a page-by-page outline so you redirect emphasis, density, or motion intent before a single slide is authored. Once you approve, Drever generates a complete story across audience view, speaker notes, document, web, and PDF from one source. The MDX and React-based authoring pipeline is deterministic and inspectable, which matters when you need to trace why a transition behaves a certain way. The wall appears early for teams who do not write MDX: this is not a drag-and-drop canvas, and non-technical stakeholders will need a developer in the loop to do anything beyond the AI-generated draft. Requires Node.js 24.18+ and a Safari or Chromium-family browser — that narrows the install base before you start.
FreeOpen SourceSelf-hostedVerified Jul 28, 2026
2. GenUI
GenUI is a native Swift workspace where agents produce declarative messages describing UI components, the client validates those messages against an allowed catalog, and SwiftUI renders only what passed validation — no agent-generated code executes on the client. The offline component Playground lets you exercise the package stack locally without standing up a hosted dependency. The renderer and local demos work; the vendor's README explicitly states the hosted gateway lacks production readiness. For teams building greenfield iOS or macOS apps that need agent-driven surfaces, the local layer is usable. For anything requiring a production-grade hosted routing layer, that piece is not finished.
FreeOpen SourceAPISelf-hostedVerified Jul 12, 2026
3. Framesmith
Framesmith is an open-source MCP server that gives AI assistants a visual design canvas by rendering HTML/CSS scene graphs to PNG via headless Chromium. The core workflow: an AI assistant sends a scene description, Framesmith processes the scene graph, and returns a PNG — no display required, no browser session to manage. The vendor states it handles component structures, custom fonts, backdrop filters, animations, and chart rendering, which covers a real slice of what design-adjacent AI workflows need. Self-hosting means the rendering pipeline stays inside your infrastructure. The ceiling appears when your workflow demands anything beyond static or lightly animated PNG output — video, interactive SVG, or live canvas — at which point Framesmith is not the right layer.
FreeOpen SourceSelf-hostedVerified Jul 21, 2026
4. Image-to-font-extractor
Feed the CLI an image and a character-order string and it produces a TTF draft, SVG glyphs, a manifest, a trace report, a contact sheet, and a browser preview — everything you need to inspect and install the result. The self-hosted Node package runs locally with no API dependency, so the full pipeline stays in your environment. Where it earns its keep is rapid prototype display fonts and logo lettering experiments, not production body text. Glyph tracing from raster sources carries inherent quality ceilings: curves traced from pixels will need manual cleanup before anything ships to a print or branding deliverable. The vendor's README explicitly flags the codebase as an AI-assisted prototype with potential dead code and magic numbers — audit accordingly.
FreeOpen SourceSelf-hostedVerified Jun 11, 2026
5. Locofy: design-to-code agents
The core workflow is plugin-based: a designer tags layers inside Figma or Penpot, configures component boundaries, and Locofy generates code the vendor describes as developer-friendly and ready to drop into a project. For straightforward landing pages, marketing sites, or mobile screens with predictable component hierarchies, teams report cutting the translation step from days to hours. The ceiling appears when designs carry complex interactive states, deeply nested conditionals, or design systems with heavy token logic — the generated code requires meaningful cleanup before it merges. Teams at that complexity level typically treat Locofy output as a scaffold rather than a final artifact, and maintain a review pass before the code reaches the repository.
PaidFree Trial · 7 days$40/moSelf-hostedVerified Jun 19, 2026
6. Pixal3d.ai
The tool accepts a single image and returns a GLB file — no multi-view capture, no turntable shoot. It runs two parallel generation lanes: the Pixal3D back-projection path and a Trellis 2 alternative, so you can compare both outputs before committing cleanup time to either. Quality presets control texture resolution (up to 2048) and vertex budget (up to 200,000 targets), which means the output ceiling is high enough for final review, not just shape prototyping. Every generation consumes credits; there is no free tier visible in the interface — you authenticate, spend credits, and download. The research weights are available on GitHub under TencentARC, so teams with the infrastructure to run inference locally are not locked to the hosted service.
Paid$9.99/moSelf-hostedVerified May 16, 2026
7. Presenton
Presenton is an open-source AI presentation generator built for the teams that cannot, or will not, route slide content through a third-party cloud. You bring a PPTX or PDF as a template, point it at your LLM of choice, and it generates full decks that inherit your colors, fonts, and layout — exported as editable PPTX or PDF. The API is the core value proposition for developers: one endpoint to generate or update a deck from your data pipeline. The visual editor covers prompt-based editing and slide variants, but the docs describe it as lacking the elaborate editing controls designers expect. Teams hitting that ceiling handle final polish in PowerPoint or Google Slides after generation.
PaidOpen Source$25.00/moAPISelf-hostedVerified Jun 9, 2026
8. Taste Lab
Point Taste at a URL and four chained agents work through the page in sequence: one pulls raw measurements across 20 categories, the next finds system-level rules, the third infers deliberate trade-offs, and a final 'observer' agent runs an anti-slop filter before writing output. The result is two files — a .md and a .json — that carry both the token set and four named design principles, each with a trigger, decision, reason, evidence, and trade-off. This makes it useful when you need a coding agent to replicate a visual language rather than just match numbers. The ceiling appears when the site's design logic is implicit or inconsistent — the pipeline infers intent, and on a messy codebase that inference will be wrong.
FreeOpen SourceSelf-hostedVerified Jun 21, 2026
Listings on this page are sourced and verified by the AIDiveForge data pipeline. AIDiveForge is editorially independent — no money changes hands for inclusion.