Breaking

ByteDance Open-Sources DeerFlow 2.0 as a Self-Hostable SuperAgent Harness

September 7, 2026 at 01:46 EDT

AI Agents
Open Source

ByteDance has released DeerFlow 2.0, a MIT-licensed open-source SuperAgent harness that coordinates language models, tools, sandboxes, memory, and sub-agents for tasks lasting minutes to hours. It is an execution runtime rather than a ready-made AI employee: users still provide the model provider, credentials, and infrastructure.

Feb. 28, 2026 · ByteDance

DeerFlow 2.0 is an open, self-hosted AI work runtime

ByteDance's open-source harness coordinates models, tools, sandboxes, memory, and sub-agents for multi-step work. It can schedule background tasks, but it is a deployable runtime rather than a free standalone employee.

Capacity check

Long-running use needs more headroom

The documented starting point doubles CPU capacity from local evaluation to a persistent server.

4 vCPU
Local evaluation

8 vCPU
Long-running server

Core flow

From prompt to deliverable

Prompt
Set the goal

→

Lead agent
Plan the work

→

Tools + sub-agents
Run specialist steps

→

Artifacts
Return usable output

The trade-off

Powerful, but not turnkey

What works

Real code execution in sandboxes
Self-hosted model flexibility
Research, coding, and content workflows

What remains hard

Docker, YAML, and CLI setup
Memory and multi-day continuity
No published independent sandbox audit

Bottom line

Open source is not the same as zero cost

MIT covers the code. Model APIs, compute, storage, monitoring, and security controls still belong to the operator.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.