Spine AI has released "Spine Swarm," a multi-agent system designed for complex non-coding projects such as competitive analysis and financial modeling. The company stated that chat interfaces are too linear and unsuitable for complex AI tasks, and has adopted a visual canvas as an alternative. This interface is characterized by making the structure of the work explicit and giving the user control.
Product LaunchesSpine AISpine Swarm
Spine AI Releases Spine Swarm, a System Where AI Agents Collaborate on an Infinite Visual Canvas
This article is a translation. Read the Japanese original
On the canvas, LLM calls, image generation, and web browsing are treated as blocks. Each block has an independent function but can be combined arbitrarily, featuring a model-agnostic design. For example, a workflow can connect an OpenAI LLM to image generation and then to app generation via Claude, utilizing different models within a single process. It is also possible to branch a single input into multiple blocks and integrate results analyzed by different models.
While the initial version required users to manually input prompts and create connections, a new chat layer has been added as a feature, allowing agents to autonomously generate and connect blocks. It has been confirmed that agents can operate autonomously over several hours to produce complete deliverables. Agents are assigned subtasks by a central orchestrator and achieve stable processing over long durations by saving intermediate context through blocks on the canvas.
Source: Launch HN: Spine Swarm (YC S23) – AI agents that collaborate on a visual canvas (HN 109pt, 69 comments) (HN Search (backfill), 2026-03-13)