Agent Weave

AgentWeave

[English](README.md) | [简体中文](README.zh-CN.md)

AgentWeave is a local-first, spatial workspace for running individual AI agents and coordinating multi-agent teams. It combines a visual canvas, durable conversations, role presets, team orchestration, and an Electron desktop shell in one pnpm monorepo.

AgentWeave is currently under active development. APIs and persisted data formats may change before the first stable release.

Product Preview

Spatial multi-agent canvas

Place agents from different providers on one infinite canvas, keep each conversation independent, and arrange the workspace around the task.


Live agent conversations

Work with an agent directly inside a resizable canvas node, including model controls, streamed responses, reasoning, tool calls, and workspace context.


Agent Team orchestration

Coordinate a leader and multiple teammates in one team node, while inspecting members, tasks, and activity from the team panel.


Reusable role presets

Create persisted roles with provider-specific instructions, then assign them consistently when building new agent teams.


Highlights

• Spatial canvas: arrange agents, teams, files, and working context on a persistent tldraw canvas.
• Live agent conversations: stream assistant output, reasoning, tools, permissions, and usage updates over SSE.
• Durable sessions: persist conversations, completed content blocks, runs, canvas snapshots, and team state in SQLite.
• Agent Teams: create a leader and teammates with isolated conversations, shared workspace context, tasks, mailbox messages, and approval-controlled spawning.
• Role presets: define reusabl