Coordinator-plus-specialists pattern, with a researcher that fetches sources, a fact-checker that verifies claims, and a writer that produces the final output. Citations are inline, and the synthesis runs on a premium model while the grunt work stays cheap.
1modules:2 web: { config: { search_backend: duckduckgo } }3 memory: { config: { working_memory: true } }4 agent_spawn: {}56agents:7 - id: coordinator8 role: coordinator9 modules: [{agent_spawn: [agent]}]10 brain: { provider: anthropic, model: claude-sonnet-5, credential: { ref: anthropic_main, scope: per_user, provider: anthropic } }11 system_prompt: |12 Spawn researcher and fact_checker via agent(agent="researcher", task="...").13 - id: researcher14 role: specialist15 modules: [{web: [search, fetch]}, {memory: [remember]}]16 brain: { provider: deepseek, model: deepseek-chat, temperature: 0.2, credential: { ref: deepseek_main, scope: per_user, provider: deepseek } }17 - id: fact_checker18 role: specialist19 modules: [{web: [search]}]20 brain: { provider: deepseek, model: deepseek-chat, temperature: 0, credential: { ref: deepseek_main, scope: per_user, provider: deepseek } }# 1. install runtime
curl -sSL https://digitorn.ai/install | sh
# 2. save the YAML above to ~/.digitorn/apps/my-research-agent/app.yaml
mkdir -p ~/.digitorn/apps/my-research-agent
# paste the YAML into app.yaml
# 3. deploy
digitorn deploy my-research-agentEngineering notes from the Digitorn team. No marketing, no launch announcements, no "10 prompts that will change your life". Just the things we write that we'd want to read.