AI NewsOpen sourceAnnouncement

D2 open-sourced its paid TALA diagram layout engine under MPL-2.0

TALA, the layout engine that was Terrastruct's paid product and funded the open-source D2 diagramming language, has been open-sourced under MPL-2.0 and merged into the D2 repository.

AI News

Editorial2 min read

LinkedInX
D2 documentation site social card

Image: D2

Why it mattersTeams that use D2 for architecture diagrams now get its best layout algorithm at no cost, and agents that generate diagrams can fix node positions while TALA handles the routing.

TALA, short for Terrastruct's AutoLayout Algorithm, is now open source under MPL-2.0, the same licence as the D2 diagramming language. D2 creator Alexander Wang announced this on the D2 blog on 7 September 2026, and the code was committed to the d2lang/d2 repository the same day, in a commit titled "open source TALA".

TALA had been the closed-source companion to D2. Wang wrote two days earlier that the development of open-source D2 had been funded by two paid products: the D2 Studio IDE and the TALA layout algorithm. The company behind them is shutting down, and D2 is continuing as a non-profit project fiscally sponsored by Hack Club. Releasing TALA is one step in that transition, and Wang says D2 Studio will follow.

What TALA does that the built-in layouts do not

D2 already included two layout algorithms, Dagre and ELK, both directional graph engines that grow diagrams in one direction. TALA is different: Wang describes it as an orthogonal layout engine designed for software architecture diagrams, closer to what people draw on whiteboards than to a top-down tree. It borrows from several graph-drawing research papers cited in the source and adds original techniques, and it weighs multiple aesthetics including symmetry, median distance, flow, and clustering of similar nodes.

The post includes seven real-world comparisons against Dagre and ELK on public D2 files pulled from GitHub, and Wang notes plainly that for some diagrams a reader may still prefer the non-TALA output. There is a runtime cost: TALA scales non-linearly, so it takes longer than the other two on large diagrams, and Wang links a benchmark repository at d2lang/d2-benchmarks for the exact numbers.

The angle for agents that draw diagrams

TALA supports fixed and hybrid positioning: some nodes can have fixed coordinates while others are left to the algorithm. Wang points this out as useful for agentic use cases, where a model can place nodes in 2D space but still struggles with routing the connectors between them. An agent that generates architecture diagrams can now describe layout intent, hand it to TALA, and let the algorithm draw the edges without paying for a proprietary component.

For a team that has been rendering D2 diagrams in a build step or a docs site, the change affects only the licence and needs no migration: the algorithm is in the same repository as the language it was built for. The paid product is gone; the code and the layout stay, funded by donations to the non-profit rather than by a licence fee.

Source

SourceD2 Blog

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project