Skip to content

Getting Started

  1. How to install the CLI
  2. Configure your API keys
  3. Run your first swarm
  4. Understand the output
  5. Next steps

Terminal window
pip install sverm

Or with Homebrew:

Terminal window
brew install sverm

Verify it’s installed:

Terminal window
sverm --version

Create a .env file in your project root (or use environment variables):

# Minimum: one provider
DEEPSEEK_API_KEY=sk-your-key-here
# Optional: add more for cost routing
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...

Sverm will auto-detect which providers you have configured and route tasks accordingly.


Terminal window
sverm run research "What is the current state of carbon capture in Norway?"

This launches a research swarm: a manager agent breaks down the question, 3 researcher agents search in parallel, a fact-checker verifies the findings, and a writer produces a report — all in 15-45 seconds.

You’ll see a live view of what each agent is doing:

[manager] Breaking down question into 3 subtopics...
[researcher-1] Searching for: Norwegian carbon capture projects
[researcher-2] Searching for: CCS technology status 2026
[researcher-3] Searching for: Norwegian CCS policy and regulation
[fact_checker] Verifying 12 claims across 3 research results...
[writer] Writing report in Norwegian...

When the swarm finishes, you’ll get:

  1. The result — a report, code review, debate transcript, or whatever you asked for
  2. A cost breakdown — which models were used, how many tokens, total cost
  3. Agent logs — what every agent did, if you want to inspect

Example cost report:

Swarm: research-swarm-v1
Total cost: $0.07
manager deepseek-v3 2,000 tokens $0.003
researcher-1 deepseek-v3 5,000 tokens $0.018
researcher-2 deepseek-v3 4,500 tokens $0.016
researcher-3 deepseek-v3 5,200 tokens $0.019
fact_checker gpt-4o-mini 1,500 tokens $0.001
writer deepseek-v3 3,000 tokens $0.008
Time: 22.3s