Spine Swarm: Multi-Agent AI System on Visual Canvas for Non-Coding Projects

Spine Swarm is a multi-agent AI system that operates on an infinite visual canvas designed for complex non-coding projects. The founders argue that chat interfaces are inadequate for complex AI work because they're linear, while real projects aren't linear. They built a workspace where work structure is explicit and user-controllable.

Core Architecture

The system uses blocks as abstractions on top of AI models. There are dedicated block types for:

• LLM calls
• Image generation
• Web browsing
• Apps
• Slides
• Spreadsheets

Blocks can be connected to any other block, with connections guaranteeing context passing regardless of block type. The system is model-agnostic, allowing workflows to move between different AI models within a single project.

Agent Operation

When a user submits a task, a central orchestrator decomposes it into subtasks and delegates each to specialized persona agents. These agents:

• Operate on canvas blocks
• Can override default settings (model and prompt) for each subtask
• Pick the best model for each block
• Sometimes run the same block with multiple models to compare outputs
• Work in parallel when subtasks don't have dependencies

Agents can pause execution to ask for user clarification or feedback before continuing. Once agents produce output, users can select a subset of blocks and iterate on them through chat without rerunning the entire workflow.

Technical Advantages

The canvas provides agents with a persistent, structured representation of the entire project that any agent can read and contribute to at any point. This addresses context degradation issues in typical multi-agent systems by:

• Storing intermediary results in blocks rather than holding everything in memory
• Creating explicit structured handoffs designed for consumption by other agents
• Allowing agents to run longer while keeping context windows clean

Users can dispatch multiple tasks at once, and the system will queue dependent ones or start independent ones immediately.