Stop drawing the graph: reactive agents over typed, versioned artifacts

Description and use

The reactifact runtime lets agents declare typed artifacts they consume and produce, automatically wiring them together without manual graph drawing, simplifying maintenance and providing provenance.

Problem

Knowledge‑agent queries often require multiple data sources (Confluence, GitLab, CSVs) and manual orchestration; maintaining a hand‑drawn graph of agent edges becomes brittle as question paths change.

Outcome

By declaring artifact types and consumption/production contracts, the system automatically composes agents without explicit graph wiring, reducing maintenance overhead and improving provenance tracking.

Workflow

• Define typed artifact models (Question, Evidence, Answer).
• Decorate agent functions with @produce to declare produced artifacts.
• Create agents with consumes and produces lists (e.g., search_agent consumes Question, produces Evidence).
• Instantiate Runtime with agents and budget.
• Create initial Question artifact in Context.
• Run the runtime; agents react automatically based on artifact availability.
• Retrieve Answer and related Evidence from Context for provenance.
• Print or use the answer and evidence.

Evidence

You don't draw a path from A to B. You declare, per agent, what it consumes and what it produces — typed artifacts, not string blobs in a shared dict.

Source

search_agent and answer_agent both react — nobody wired them together

Source