You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The LangGraph-equivalent core: a directed graph of nodes and edges over a shared, typed run state. Executes a graph to completion, threading state between nodes. Foundation for everything in P1–P3.
Scope
Define the graph/state data model (nodes, edges, typed shared state)
Topological / scheduler-driven execution over the graph
State reducers / channels so nodes read and update shared state safely
Deterministic execution given the same inputs + state
Acceptance criteria
A graph of ≥3 nodes executes end-to-end with state passed between them
State updates are isolated per-node and merged via defined reducers
Unit tests cover linear, fan-out and fan-in graphs
I would like to work on this! I have an implementation ready that covers the complete control-flow specification:
Conditional / router edges with safe AST expression evaluation and callable predicates
Parallel branch fan-out and join (barrier synchronization) with isolated state copies and reducer merging
Bounded loops / cycles with termination guards and configurable iteration limits
Cycle detection that distinguishes between valid guarded loops and malformed infinite cycles
Since #278 builds directly on the graph/state runtime core (#277), I've implemented both in a clean dynavec.agents module with zero regressions to the existing ReActAgent/Planner primitives, accompanied by a comprehensive offline test suite.
Could you please assign this to me? Let me know if you would like this submitted as a combined dynaflow engine PR (#277 + #278) or split into two sequential PRs.
Activity
Ashishds commentedon Oct 7, 2026
Hi @codeforstartups,
I would like to work on this! I have an implementation ready that covers the complete control-flow specification:
Since #278 builds directly on the graph/state runtime core (#277), I've implemented both in a clean
dynavec.agentsmodule with zero regressions to the existing ReActAgent/Planner primitives, accompanied by a comprehensive offline test suite.Could you please assign this to me? Let me know if you would like this submitted as a combined dynaflow engine PR (#277 + #278) or split into two sequential PRs.
feat(agents): add dynaflow graph orchestration engine and control flow (
feat(agents): add dynaflow graph orchestration engine and control flow (
Kaap10 commentedon Oct 13, 2026
Resolved by #344.