Prose-like DSL runtime with OpenAI execution. Parse, validate, and run agent workflows (parallel blocks, loops, conditionals, retries, pipelines).
Docs:
docs/usage.mddocs/language.mddocs/runtime.md
bun installexport OPENAI_API_KEY="your-key"
bun run index.ts run workflow.prose --model gpt-4o-minibun run index.ts validate workflow.prose
bun run index.ts parse workflow.proseagent researcher:
model: "gpt-4o-mini"
prompt: "You are a research assistant"
agent writer:
model: "gpt-4o-mini"
persist: true
parallel:
research = session: researcher
prompt: "Research quantum computing breakthroughs"
competitive = session: researcher
prompt: "Analyze competitor landscape"
loop until **the draft meets publication standards** (max: 3):
session: writer
prompt: "Write and refine the article"
context: { research, competitive }
Persistent agents store memory across runs. Use persist: true on the agent and resume: agentName in the workflow:
agent writer:
model: "gpt-4o-mini"
persist: true
session: writer
prompt: "First draft"
resume: writer
prompt: "Revise with the prior memory"
Run artifacts are written to .prose/runs/<runId>/ by default (state, bindings, agent logs). Persistent agent memory is stored in ~/.prose/agents/.
You can control persistence via CLI flags:
bun run index.ts run workflow.prose --no-persist
bun run index.ts run workflow.prose --runs-dir ./my-runs --agents-dir ./my-agentsThis project was created using bun init in bun v1.3.5. Bun is a fast all-in-one JavaScript runtime.