featured · github
Deer Flow: ByteDance's Open Long-Horizon Agent Framework
A multi-agent orchestration layer that handles research-to-code-to-creation workflows in a single harness. Useful when one AI call isn't enough.
bytedance/deer-flow ↗Deer Flow is a scaffolding system for building agents that tackle multi-step projects over minutes or hours. Instead of one AI model doing one thing, it coordinates multiple specialized sub-agents (one researches, one codes, one executes), syncs their outputs through memory, runs them in isolated sandboxes, and gates access to tools. Think of it as a project manager for AI workers. The payoff: complex workflows that would normally require manual orchestration glue code now have a declarative framework. Builders get built-in task memory, subagent delegation, and message routing. It's not a magic black box—you're still engineering the flow—but it removes the plumbing overhead for long-running, multi-stage tasks.
Share kit
Deer Flow: Long-horizon agent coordination, now open source
ByteDance released Deer Flow, an orchestration framework for multi-step AI workflows. It handles research-to-code-to-create pipelines with subagent delegation, memory, sandboxed execution, and tool gating. If you're wrangling multi-agent workflows today, this cuts implementation friction. GitHub link in the report.
ByteDance open-sourced Deer Flow—a framework for long-horizon agent workflows that chain research, coding, and creation in a single harness. Subagent coordination, memory, sandboxed execution. If you're stitching multi-agent systems together manually, this is worth a look. https://github.com/bytedance/deer-flow
Shipping complex AI workflows? ByteDance just open-sourced Deer Flow—an orchestration framework that coordinates multiple specialized agents (research, code, create) in a single long-horizon task. Built-in memory, sandboxing, and tool gating. Cuts the plumbing overhead for multi-step AI projects. Worth exploring if you're building at scale.
Just found ByteDance's deer-flow and it's the clearest take on long-horizon task orchestration I've seen in a while. Most agent frameworks stop at single-turn reasoning. This one handles research, coding, and creation workflows that span minutes to hours. The real win: sandboxes, memory, subagents, and a message gateway all built in from the start, not bolted on. I've watched teams build this exact stack three times over. Having it open-sourced and battle-tested at scale changes the baseline for what you can ship. If you're orchestrating multi-step workflows with AI, worth a look. https://github.com/bytedance/deer-flow
deer-flow from ByteDance is a solid open-source framework for long-horizon agent workflows. Handles research→code→create pipelines with sandboxes, memory, subagents, and a message gateway baked in. Most tools stop at single turns. This one goes to completion. https://github.com/bytedance/deer-flow