Spine Research

Spine Research spins up specialized agents across 300+ AI models and runs their work in parallel. You describe a project in plain language and agents plan, decompose and execute on the canvas, creating research, document and data blocks as they go. That parallelism is what lets research automation that would take days of manual sourcing land in hours.

The canvas is an infinite workspace where you and the agents both work. You add blocks, connect the context they should read and branch in different directions from one starting point. Because the structure is visible, you can let agents do the heavy lifting and then pick a subset of blocks to refine by hand without rerunning the entire workflow.

The Deep Research block runs models in a loop behind the scenes: a planning model decides what and where to search, a writing model consolidates the findings. It browses the web and any documents on your canvas, supports include and exclude rules for domains and date ranges, and attaches source citations so you can check the claims.

Output lands as work products, not walls of text: reports, memos, slide decks, spreadsheets and data tables. Every run keeps its sources and reasoning attached, so the deliverable carries its evidence. You can edit any piece directly and share the whole canvas or a single artifact by URL.

Before a task runs, plan mode shows you the agent's proposed approach so you can adjust it, ask for something simpler or trim the scope. You can also feed documents, links and websites onto the canvas so agents read that context directly instead of searching for it, which keeps both cost and guesswork down.

Beyond the visual app, Spine Research offers a research API that returns polished artifacts from a single call, with webhooks and polling for async runs. It also plugs into tools like Claude Code, Codex CLI, Cursor and OpenClaw, so agents in those environments can delegate research and deliverable generation back to Spine.