01. Product Overview

Heicode is an intelligent development tool that covers the full software lifecycle. It targets the complete "idea to production" development process, not a single model chat window, the CodeGW admin backend, or a simple Agent console.

Users only need to register, log in, and describe their idea and constraints — Heicode then organizes a team of sub-Agents, connects to the Git, SK, project docs, and cloud resources the user authorizes, and carries out product design, coding, code review, deployment, observability, and ongoing maintenance.

Current status: what's currently connected and production-verified is the template Agent main path (a user picks a template in Manager, deploys a persistent Agent with resources, connects directly from the client, and model calls are billed through Manager). Multi-Agent swarm runtime lives in agent_swarm (HeiCode Swarm), pending the runtime contract being frozen before full integration. The complete "idea to production" lifecycle loop is still under construction — this document describes the target product shape, not that every part is fully shipped.

Target Users

User	Typical Problem	Heicode's Value
Independent developer	Has an idea but no engineering team	Use an AI team to go from requirements to production
Startup team	Small headcount, fast iteration, delivery pressure	Use Heicode to orchestrate product, dev, test, and deploy workflows
Enterprise innovation team	Has resources and code, but collaboration overhead is high	Controlled development within existing Git and cloud resource boundaries
Tech lead	Worried about permissions, cost, quality, and audit	Unified resource authorization, model usage, logs, and audit

Core Value

Turns a natural-language idea into an executable software development task.
Turns Git, docs, SK, and cloud resources into task context the user can connect, confirm, and revoke through a guided flow.
Automatically organizes sub-Agent roles, permissions, models, and context into a working dev team — ordinary users only confirm a summary.
Breaks ongoing development into requirements, design, coding, testing, fixes, and deployment sub-stages, driven forward by Agents.
Lets Agents call project SK tools and external capabilities during execution, not just a one-shot task dispatch.
Presents model calls, balance, quota, and logs to the user in one place.
Puts long-lived credentials in a credential vault, giving sub-Agents only short-lived, least-privilege credentials.
Folds development, review, deployment, and maintenance into the same lifecycle loop.

Product Surfaces

Surface	Description
Heicode Client	The user's main experience — local conversation, entering ideas, continuing development, viewing execution feedback, receiving deliverables, and approving high-risk actions. Only logs into Heicode, and only uses models Heicode provides.
Heicode Manager / browser console	An auxiliary console — handles account & security, client downloads, the resource-preparation wizard, Agent startup, task status overview, model balance & usage, and audit/log viewing.
Agent Platform	The execution layer — deploys and runs sub-Agents, carries requirements, coding, testing, fixes, delivery, and deployment forward during a task, and reports back logs, status, events, and metrics.
CodeGW	The internal model gateway and billing service; ordinary users don't access its backend directly.
Credential vault	Implemented on OpenBao; stores long-lived credentials and issues short-lived credential leases based on approval and permissions.

Note: the Heicode client is the user's primary experience — users don't code in a browser. Heicode Manager is the auxiliary browser console, handling resource prep, Agent startup, status, balance, audit, and downloads. Manager shouldn't confront users with a full parameter surface — it should guide them through a task-readiness checklist toward the confirmations that actually matter. What actually keeps a task moving forward is the Agent execution loop that Heicode orchestrates; during execution, Agents can also call authorized SK tools and external capabilities, ultimately completing delivery and deployment and reporting results back to the client.

Core Flow

Register & log in
-> Enter a product idea
-> Generate a product doc and task plan
-> Connect Git, SK, docs, and cloud accounts in Manager via the readiness checklist
-> Heicode auto-discovers resources and generates an execution summary
-> User confirms the recommended scope and starts the Agent
-> Client keeps adding requirements and steering direction
-> Agent works through requirements, dev, test, and fixes by sub-stage
-> Agent calls authorized SK tools as needed
-> Client approves high-risk actions
-> Agent finishes packaging deliverables and deploys
-> Manager shows status, balance, audit, and downloads
-> Client shows execution feedback and deliverables
-> Ongoing maintenance and upgrades

What Makes Heicode Different

Compared to	Difference
Ordinary AI chat tools	Heicode doesn't just answer questions — it organizes resources, permissions, and an execution team
IDE plugins	Heicode isn't limited to local code editing — it covers deployment, audit, and the full lifecycle
The CodeGW backend	Heicode doesn't manage a channel backend — it's user-facing, focused on getting development tasks done
Traditional DevOps platforms	Heicode is built around natural language and an AI team, auto-generating tasks and execution context

Product Commitments

Users don't need to understand the underlying model provider.
Users don't need to hand credentials to sub-Agents for long-term storage.
Users don't need to view or fill in full technical parameters — only confirm the resources, risks, and execution summary the system has organized.
High-risk actions must go through client-side approval.
Every key action is auditable.

Starting from an idea, let Heicode organize an AI development team that keeps calling on Agents to develop, test, deliver, and deploy — securely connecting to your code and cloud resources — and take your product to production.

Introduction