DeerFlow on Railway
ByteDance's open-source research agent: plans, searches, delegates and writes cited reports, on Railway.
DeerFlow is an open-source agent harness from ByteDance. You give it a goal and a lead agent plans the work, searches the web, reads sources, delegates to sub-agents and returns a cited report or other files. It is used for deep research, literature reviews and recurring research digests.
Version 2 adds custom agents, schedules, skills and MCP tools. The comparison people usually make is with hosted deep-research products from the large AI vendors; DeerFlow differs in that you run it yourself and choose the model provider.
DeerFlow is released under the MIT licence.
What DeerFlow can do
- Lead agent that plans and delegates to sub-agents
- Web search and page fetching without a key
- Cited reports and file outputs per thread
- Custom agents, schedules and MCP tools
- Bundled skills such as academic research
- Bring your own model provider key
What people use it for
- Market, competitor and technology research with a cited briefing
- Literature reviews and paper summaries
- Recurring research digests on a schedule
- A private research assistant for a team
Videos
Running DeerFlow on Railway
The template runs the DeerFlow Gateway (agent runtime and API) and the Next.js frontend, both v2.1.0, behind an nginx edge proxy on one public URL. Postgres 18 stores users, threads and run history, Redis 8.2 carries live run events, and a volume holds each thread's files. An admin account is generated and public sign-up is closed. Cost follows the services' CPU and memory; model calls are billed by your provider.
DeerFlow templates on Railway
Tracklayer templates are listed first. Other DeerFlow templates follow, ordered by total deployments on Railway. How this works.
Deployment numbers as of 2026-11-20.
Questions about DeerFlow
How do I log in?
Wait until the Gateway logs that the admin account was created, open the Nginx URL and sign in with DEERFLOW_ADMIN_EMAIL and DEERFLOW_ADMIN_PASSWORD from the Gateway variables. Public sign-up is off.
Which model providers work?
Set one of ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, DEEPSEEK_API_KEY or OPENROUTER_API_KEY on the Gateway and redeploy. The provider bills you directly.
Can agents run code?
Not by default. Railway has no Docker-in-Docker, so DeerFlow's container sandbox cannot run. Set E2B_API_KEY to run code in E2B cloud sandboxes.
Does web search need a key?
No. Search uses DuckDuckGo without a key. TAVILY_API_KEY switches to Tavily and JINA_API_KEY raises page-fetch limits.
What drives the cost?
The Gateway's CPU and memory during concurrent runs, plus Postgres, Redis and the volume. Run the Gateway as one replica and size it up for more simultaneous runs.
Similar apps
Tracklayer maintains the templates marked as such; DeerFlow is a trademark of its owners and this site is not affiliated with them.