BatonBot / LocalChainAI

A Chat-Driven App for software building: user requests flow through an LLM that emits structured task instructions, which BatonBot’s orchestrator parses and dispatches to CLI coding agents in parallel.

Chat-Driven AppAI OrchestrationNode.jsMulti-AgentLocal-firstLM StudioAnthropic

Visit www.batonbot.io →

Key wins

▸Processed ~128M tokens in 43 minutes from a single user workflow
▸Runs multiple isolated AI build sessions in parallel
▸Skill-file interface for external agents and tools

As a Chat-Driven App

BatonBot uses the Chat-Driven App pattern to turn messy human input into structured pipeline jobs. One of its features: a sanitizer agent that cleans up a long, freeform todo list into discrete tasks the orchestrator can dispatch.

User input

Here’s my dump of stuff to do today, clean it up and run it: fix login bug. also the readme is wrong. write tests for the parser maybe. deploy preview???

LLM output

[ACTION:enqueue_task title="Fix login bug" priority="high"]
[ACTION:enqueue_task title="Correct README inaccuracies" priority="medium"]
[ACTION:enqueue_task title="Write unit tests for parser module" priority="medium"]
[ACTION:enqueue_task title="Deploy preview environment" priority="low"]

What the app does

BatonBot parses each [ACTION:enqueue_task] tag and pushes the cleaned task into its orchestration pipeline, where each task is routed to an isolated CLI coding-agent session with its own logs and result stream.

Read about the Chat-Driven Apps pattern →

What I built

Built an end-to-end macOS prototype that orchestrates AI-assisted software workflows across local and cloud model providers, including LM Studio, OpenAI-compatible endpoints, Grok, and Anthropic.
Developed a Node.js-based orchestration layer that routes tasks, captures logs, coordinates multi-step execution logic, and runs multiple isolated AI build sessions in parallel.
Designed skill-file interfaces that allow external agents and tools to consume reusable workflow instructions and participate in BatonBot-managed task execution.
Integrated local AI development workflows using LM Studio, Cline, Aider, OpenClaw, VS Code agent extensions, and CLI coding agents to accelerate coding, debugging, orchestration, and test iteration.
Designed workflows that convert requirements into testable software outputs with human-reviewable logs and repeatable execution paths.
Demonstrated high-throughput parallel workflow execution, including a recorded test processing approximately 128M tokens in 43 minutes from a single user workflow.

← Back to all projects