Heicode goes fully live on June 5, 2026.
Register now and follow launch readiness for Client, Manager, Agent Swarm, resource binding and audit.
One idea,one AI development team.
Heicode organizes sub-Agent teams, connects your code, documents, SK skills and cloud resources, and continuously carries work through requirements, development, testing, fixes, deployment and maintenance.
Heicode Task Cockpit
Small-team task management SaaS
User input
I want to build a small-team task management SaaS with login, projects, tasks, comments, notifications and admin, deployed to Azure.
Product Agent
Requirements and acceptance
Done
Backend Agent
APIs and data model
Building
Frontend Agent
Pages and interaction
Waiting for API
Ops Agent
Deployment and observability
Approval pending
Production deployment requires client approval
The secret vault will derive a 15-minute short-lived credential. Sub Agents never receive long-term secrets.
User journey
Users see one continuous flow, not a wall of admin panels
The homepage and primary workflow stay anchored to the active job: idea, resources, team, permissions, execution, launch, and maintenance.
Minimum MVP path
Sign in, download the client, describe the idea, bind Git and cloud resources, confirm role permissions, preview the manifest, create an Agent deployment placeholder, then keep development and testing moving.
Sign in
Users only sign in to Heicode, not CodeGW or the OpenBao console.
Describe the idea
Capture the product, constraints, existing code, and deployment target in the client.
Bind resources
Manager authorizes Git, SK, docs, and cloud accounts while storing secret_ref only.
Generate the team
Recommend Product, Architect, Frontend, Backend, Reviewer, Ops, and other sub Agents.
Assign permissions
Confirm resources, actions, TTL, and approval requirements through role cards.
Close the loop
Agents keep pushing requirements, development, testing, fixes, deployment, and maintenance.
Users do not write manifests
The default flow shows roles, resources, actions, and approvals. Advanced users can expand the permission manifest.
The web app is not the IDE
Manager handles auxiliary management and observation. Main development, continuation, and approvals return to the client.
Risky actions need explicit approval
Production deploys, cloud operations, database writes, and production secret access must be confirmed in the client.
Product surfaces
Users operate Heicode. The underlying systems stay in their lanes.
The client is the primary experience and Manager is the auxiliary console. CodeGW, the vault, and the Agent platform are not exposed as ordinary user backends.
Heicode Client
The primary user experience. Describe ideas, continue development, review Agent execution feedback, and approve risky operations before production deploys, secret access, or cloud changes.
Heicode Manager
The browser-side control console for account safety, client downloads, resource binding, Agent deployment, task status, balance, and audit.
Agent Platform
The AI development-team execution layer. It runs Product, Frontend, Backend, Reviewer, Ops, and other sub Agents on AKS, then reports state and events back.
CodeGW
The internal model gateway and billing foundation. Heicode only exposes the models, balance, quota, usage, and call logs ordinary users need.
Secret Vault
Credential custody backed by OpenBao. Long-lived secrets only enter the vault; sub Agents receive short-lived, least-privilege credentials.
Resource Context
Git, SK, documents, cloud accounts, and cloud resources become Resource Bindings and Grants that can be authorized, revoked, and audited.
Core abilities
Four loops Heicode must close
From idea, resources, team, and permissions to execution, launch, and maintenance, Heicode turns software delivery into an auditable lifecycle.
Ideas become development work
Users do not need a complete PRD first. Heicode turns natural-language goals into requirement summaries, task drafts, prototype notes, role suggestions, and resource needs.
Resource binding and permission assignment
Git, SK, project docs, and cloud resources are modeled as Resource Bindings, then assigned to specific sub Agent roles through Resource Grants.
Sub Agents execute continuously
Work is not thrown over the wall once. Agents keep moving through requirements, design, development, testing, fixes, and deployment.
Usage, logs, security, and audit loop
Model balance, call logs, resource access, risky approvals, and deployment records return to Heicode. Long-lived secrets never enter Git, Markdown, or logs.
Security model
Resources are usable. Secrets do not leak.
Users authorize resources, the platform holds credentials, and sub Agents only receive short-lived access inside the approved scope.
Secrets never enter the product surface
Git tokens, cloud access keys, SSH private keys, database passwords, and CodeGW keys never enter Git, Markdown, frontend responses, or ordinary logs.
Only secret_ref is stored
The Manager database stores resource metadata, permission relationships, and secret_ref. Real long-lived credentials are held by the vault.
Sub Agents only receive short-lived credentials
After approval, the platform derives short-lived, least-privilege, auditable credentials by permission and TTL. Sub Agents do not keep long-lived secrets.
Risky actions are approved in the client
Production deploys, cloud changes, database writes, production secret access, and large budget consumption require explicit client confirmation.
Audit answers the critical questions
Who used which resource, for which task, through which sub Agent, at what time, with what action, and whether approval happened.
MVP to lifecycle
Prove the loop first, then expand the full lifecycle
The real Agent platform deployment API can arrive later. Manager payloads, manifests, audit trails, and the client approval loop must be ready first.
Current MVP
Heicode, CodeGW, and OpenBao deploy separately. Resource bindings, Grants, manifests, and secret custody form the first closed loop. The client signs in only to Heicode.
Next
Expose user-side CodeGW model and usage views, close the client login and approval download path, and integrate the real Agent deployment API.
Long term
Add automated maintenance, upgrade plans, and continuous cost and quality observation so development, deployment, and maintenance live in one lifecycle.
Move an idea into a real development flow
Heicode does more than generate code. It organizes product intent, code, permissions, models, deployment, and maintenance into one auditable software lifecycle.