Skip to main contentSkip to chat inputSkip to workbench
ChromaFlow
System Architecture

Multi-agent control plane

Orchestrated execution graph with parallel agent scheduling, dependency resolution, and deterministic deployment to edge runtimes.

Topology

Edge + Workers + State

Agents

6 Specialized

Latency

<50ms p99

INPUT LAYERIngress and artifactsORCHESTRATION LAYERPlanning and schedulingAGENT LAYERSpecialized parallel workersTOOLING LAYERExternal systems and APIsRUNTIME / DEPLOYMENT LAYEREdge execution and targetsNatural Language…UI / APIFile UploadsArtifactsAPI CallsProgrammaticMulti-Agent Cont…Control planeTask PlannerSpec → tasksDependency Resol…DAG builderExecution GraphSchedulerFrontend AgentRoutes / UIBackend AgentAPIs / schemaInfra AgentDeploy / runtimeAI AgentModel opsQA AgentTestsSecurity AgentPolicyGitHubRepo / PRSupabasePostgres / AuthCloudflarePages / WorkersDatabasesPostgres / KVCI/CDChecks / gatesInfrastructure A…ProvidersEdge RuntimesLow latencyWorkersComputeDurable ObjectsStateProduction TargetsDeploySIGNAL FLOWControlDataExecutingComplete

Hover or tap nodes to inspect. Click to pin selection.

Execution Semantics

  • Planner emits tasks with explicit inputs/outputs and acceptance checks.
  • Resolver materializes a DAG and identifies safe parallel regions.
  • Execution graph schedules in parallel with deterministic handoff.
  • Agents operate under scoped tool contexts.

Integration Boundaries

  • GitHub is the change ledger for code and review artifacts.
  • CI/CD gates execution with tests, checks, and deploy approvals.
  • Supabase and database primitives represent the data plane.
  • Cloudflare provides edge surfaces for runtime and deploy.

Audit & Security

  • Tool boundaries enforce least-privilege scopes per agent.
  • Execution metadata supports traceability across plan → code → deploy.
  • Policy checks run as gates (secrets, dependencies, static analysis).
  • Runtime configuration via controlled infrastructure APIs.