Heicode Docs
Overview

Heicode Overview

Understand what Heicode is, its core components, running modes, and capability boundaries.

Heicode Overview

Heicode is an AI Agent platform built for software development tasks. It brings project context, model access, tool calls, command execution, code changes, test validation, and multi-Agent collaboration together into a single workflow.

What Heicode Solves

Traditional AI coding tools usually focus on single-turn Q&A or local completion. Heicode focuses on the full task loop instead: understand the goal, read the repository, break the task down, apply changes, run tests, gather evidence, and hand the final decision back to the user.

Core Components

ComponentPrimary Role
Heicode DesktopEntry point for local projects, sessions, file changes, command approval, and viewing results
Heicode CloudAccounts, organizations, quota, task records, and the cloud control plane
Agent RuntimeExecutes single-Agent development tasks
Agent SwarmMulti-Agent task decomposition, parallel execution, review, and result aggregation
Model GatewayConnects different model providers behind a unified request entry point
MCP / ToolsProvides Agents with external tools, data sources, and system capabilities
Sandbox / WorkspaceProvides an isolated execution environment for remote tasks

Single Agent vs. Agent Swarm

A single Agent is a better fit for tasks with clear boundaries, a small change scope, and a need for fast interaction. Agent Swarm is a better fit for complex, long-running tasks that can be split up and run in parallel.

Multiple Agents are not inherently better than a single Agent. When a task is too small, its acceptance criteria are unclear, or the budget is insufficient, using a Swarm may only add cost and coordination overhead.

Running Modes

ModeCode LocationExecution LocationBest For
Local modeLocal deviceLocal devicePrivate code, local debugging, low-latency interaction
Cloud modeCloud workspaceRemote Sandbox or VMLong tasks, parallel tasks, continuous execution
Hybrid modeLocal and cloudLocal + cloudLocal control, remote execution, multi-device collaboration

What Heicode Is Not

Heicode cannot guarantee that every task completes automatically, and it cannot replace code review, testing, permission management, or release processes. Code, commands, test results, and conclusions produced by an Agent should all be verified.

Suggested Reading Order

  1. Read "First Success Path" and complete your first verifiable task.
  2. Read "Agent Permissions & Security Boundaries" and confirm which actions the Agent is allowed to perform.
  3. Read "Task Results & Acceptance" to understand the difference between completed, passed, and resolved.
  4. Before using Agent Swarm, read up on the cost, concurrency, and coordination mechanics of multiple Agents.

Last updated on

On this page