Execution Topologies

April 13, 2026 ยท View on GitHub

Qualixar OS supports 13 execution topologies that define how agents communicate and process work. These are registered in the swarm engine from four topology modules: basic (4), advanced (4), experimental (4), and hybrid (1).

Topology Quick Reference

#TopologyModuleAgentsPatternBest For
1sequentialbasic2+ChainMulti-step processing
2parallelbasic2+ConcurrentIndependent subtasks
3hierarchicalbasic3+Manager + workersComplex projects
4dagbasic2+Directed acyclic graphCustom dependencies
5mixture_of_agentsadvanced3+Ensemble + synthesizerBest-of-N synthesis
6debateadvanced3Argue + judgeCreative decisions
7meshadvanced3+All-to-allCollaborative reasoning
8staradvanced3+Hub-and-spokeCentralized coordination
9circularexperimental3+Ring passingIterative refinement
10gridexperimental4+2D grid neighborsSpatial/parallel reasoning
11forestexperimental4+Multiple tree rootsParallel hierarchies
12makerexperimental2+Build-test cycleCode generation
13hybridhybrid3+Local + cloud routingCost-optimized execution

Detailed Descriptions

Sequential (basic)

Agents execute in order. Each agent's output becomes the next agent's input.

Agent A --> Agent B --> Agent C --> Result

Use when tasks have clear sequential stages (research, write, review).

Parallel (basic)

All agents execute simultaneously on the same prompt. Results are collected independently.

Use for independent subtasks or getting multiple perspectives.

Hierarchical (basic)

A manager agent decomposes the task and assigns subtasks to worker agents. Workers report back to the manager, who assembles the final result.

DAG (basic)

Define a custom directed acyclic graph of agent dependencies. Each node executes when all its dependencies complete.

{ "topology": "dag", "topologyConfig": { "edges": [["research", "write"], ["research", "diagram"]] } }

Mixture of Agents (advanced)

Multiple agents independently produce outputs. A synthesizer agent merges the best elements into a final result. Similar to voting but with intelligent synthesis rather than simple majority.

Debate (advanced)

Two agents argue opposing positions across multiple rounds. A third agent (judge) evaluates and decides.

Mesh (advanced)

Every agent can communicate with every other agent. Messages pass around until the group converges on a result. Use for collaborative reasoning.

Star (advanced)

A central hub agent communicates with all spoke agents. Spokes do not communicate with each other. The hub coordinates and aggregates results.

Circular (experimental)

Agents are arranged in a ring. Each agent processes and passes output to the next in the circle. Multiple rounds refine the result iteratively.

Grid (experimental)

Agents are arranged in a 2D grid. Each agent communicates with its neighbors (up, down, left, right). Use for spatial reasoning or parallel processing patterns.

Forest (experimental)

Multiple independent tree hierarchies execute in parallel. Each tree has its own root and workers. Results from all trees are merged.

Maker (experimental)

A builder agent generates output (typically code), then a tester agent validates it. Failed tests trigger another build-test cycle. Use for code generation with automated verification.

Hybrid (hybrid)

Routes agents between local and cloud execution based on cost, latency, and capability requirements. The 13th topology, designed for cost-optimized mixed-provider workflows.

Emits hybrid:route_assigned, hybrid:cloud_fallback, and hybrid:cost_reconciled events.

Mode Gating

Not all topologies are available in all modes. The ModeEngine enforces topology access:

  • Companion mode: Simpler topologies allowed
  • Power mode: All 13 topologies available

Check available topologies via: GET /api/swarm/topologies

Choosing a Topology

QuestionRecommended
Simple, one-step task?sequential (1 agent)
Clear sequential stages?sequential
Independent subtasks?parallel
Need best-of-N?mixture_of_agents
Complex dependencies?dag
Exploring trade-offs?debate
Code generation?maker
Cost-sensitive multi-provider?hybrid