2.4 compatibility in Nexa 3.0

This page describes the retained 2.4 language/runtime path. Sources without a version declaration use this path. Start new Open Programs with Nexa 3.0; compare contracts in the migration guide.

Pipelines and parallel orchestration

Choose topology from data dependencies, then set budgets. Sequential pipelines pass one step's output to the next; parallel branches use isolated Agent sessions.

Sequential pipelines

Configure a model, save this as pipeline.nx, and run uv run nexa run pipeline.nx. Researcher extracts facts; Summarizer receives those facts and returns a summary.

agent Researcher { prompt: "Extract the facts." }
agent Summarizer { prompt: "Summarize the facts." }
flow main {
    result = "evidence" >> Researcher >> Summarizer;
}

Each stage receives the previous result. When both endpoints declare context, the pipeline transfers AgentContext. Explicit calls such as facts = Researcher.run("evidence"); result = Summarizer.run(facts); are useful when you need to inspect an intermediate value.

fork/merge

The following program explores two approaches in parallel and requires model configuration.

agent Explorer { prompt: "Explore one approach." }
flow main {
    fork [Explorer.run("Approach A"), Explorer.run("Approach B")] merge all;
}

Concurrency defaults to a bound of 8 and results preserve input order. all collects results; first and vote select according to their strategies. best/best_of require a numeric score in each result and fail when it is absent. Completion order is distinct from input order.

Repeated messages

Use an Actor to send successive messages to one worker. Use execute and checkpoints for cross-process approval persistence.

---

Source and validation · language / harness / context