A glowing orchestration core gathering light from cloud, GPU tower and laptop into one dew-like stream through an approval gate

Dew Flow

Multi-agent software development orchestrator
Ask about this product
AI developmentOur product

Dew Flow

Multi-agent software development orchestrator

Dew Flow is the delivery engine behind RemSoft.Dev's own client work. Built on .NET 10 and Aspire, it takes a work item from Azure DevOps, Jira or plain text and carries it through specification, implementation, tests and a pull request. A dependency-aware scheduler routes complex reasoning to Claude, Codex or Gemini and implementation tasks to local LLMs; Rust components, AMD drivers and ROCm run those models through WSL on Windows for maximum local performance. GraphRAG and code-RAG give every agent the context of the real codebase. Every plan and every diff passes ConnectOtherAIs, our independent review gate, where models from other vendors check the work before a person sees it - and CredsForDevs lets agents use servers and databases without ever receiving a password. Human approval gates keep engineers in charge, token and cost tracking makes every run accountable, and a dedicated test framework validates agent flows, model routing and system stability.

Ask about it

Features

  • Azure DevOps, Jira and free-text work items as input
  • Dependency-aware scheduler across Claude, Codex, Gemini and local LLMs
  • GraphRAG and code-RAG context from the real repository
  • ConnectOtherAIs review gate on every plan and every diff
  • CredsForDevs broker: agents act on credentials without receiving them
  • Local models on AMD GPUs through Rust, ROCm and WSL
  • One shared rule set for every agent session across repositories
  • Dedicated test framework for agent flows and model routing

The Challenge

Product: Multi-agent development orchestrator

Origin: Built for RemSoft.Dev's own client delivery

Agents: Claude Code, Codex, Gemini, local LLMs

Platform: .NET 10 / Aspire, Rust, ROCm/WSL

What the product had to solve:

  • Orchestrating unlike models: Cloud agents and local models have different strengths and failure modes - the scheduler routes work by task type and dependency, not just availability.
  • An agent cannot review itself: A model reading its own diff agrees with itself. Dew Flow sends every plan and diff to other vendors' models through ConnectOtherAIs.
  • Keeping humans in charge: Approval gates and review checkpoints are part of the architecture: nothing merges because an agent said so.
  • Local performance: Getting real throughput from local models on AMD hardware required Rust components and a ROCm/WSL execution path.

Results

  • 15 orchestrated AI agents
  • Work item to pull request under human approval
  • Cloud and local model routing
  • Independent review on every change
  • Proven on 2 commercial projects

15 agents

collaborating on planning, coding, testing and review.

2 projects

commercial deliveries validated the system.

Every change

reviewed by other vendors' models before a person sees it.

Case studies from our work

Interested in This Product?

Tell us how you would use it - we will answer questions on availability, setup and custom versions.

Ask about this product
An unhandled error has occurred. Reload 🗙