aglet-builtin-planner-workflow

Aglet declarative-DAG planner: walk a fixed node graph of tool calls instead of letting an LLM decide.

Part of the Aglet pluggable Agent runtime.

Use planner.workflow when the call graph is known up front (extract → summarise → render), and you want the determinism and predictability of a classical pipeline while still plugging into Aglet's Event / Patch / Budget / Hook / Observability stack.

Install

pip install --pre aglet-builtin-planner-workflow

Example

planner:
  techniques:
    - name: workflow
      config:
        nodes:
          - id: fetch
            tool: http_get
            arguments:
              url: "https://example.com/api/{input}"
          - id: final
            final: "Fetched status: {nodes.fetch.status}"

The planner yields one tool call per round (PLANNER_ACTION with next_action set), the Runtime dispatches it through the Executor, and the planner resumes on the next iteration with the result available at {nodes.<id>}.

Templates support:

• {input} — user's raw input text.
• {nodes.<id>} — full output of a previous node.
• {nodes.<id>.<field>} — nested-dict field access.

License

Apache-2.0