// case study
Alara Code
A secure, multi-tenant platform that unifies planning and building in one workspace: run the full ECC-SDLC pipeline (scope, requirements, design, test specs, estimates, proposals), then generate the actual code with a two-pass AI pipeline, review it in a built-in six-pane Browser IDE, and ship it as GitHub pull requests. Model-agnostic, working with Claude, GLM, and more.
01 The problem
Software teams split planning and building across disconnected tools, requirements live in docs, design in another app, code in an IDE, and AI assistants are bolted on the side. That makes it hard to:
1. Move from requirements and design straight into working, traceable code
2. Swap between AI models (Claude, GLM, and others) without rewiring everything
3. Run AI code generation safely and review it before it ships
4. Keep the whole flow multi-tenant, secure, and auditable for real clients
1. Move from requirements and design straight into working, traceable code
2. Swap between AI models (Claude, GLM, and others) without rewiring everything
3. Run AI code generation safely and review it before it ships
4. Keep the whole flow multi-tenant, secure, and auditable for real clients
02 The solution
Alara Code brings the entire lifecycle into a single platform:
1. Pre-development pipeline: run ECC-SDLC commands (scope → SRS → SDS → STS → estimate → proposal) with live output streamed over Server-Sent Events and artifacts stored in the cloud
2. Two-pass AI code generation: Pass 1 scaffolds the full project skeleton, Pass 2 implements each design component, powered by the Claude Agent SDK and model-agnostic (Claude, GLM, and more)
3. Browser IDE: a six-pane code-review and preview environment with file-level accept / reject
4. Safe execution: every generation runs in a resource-limited Docker sandbox with network allow-listing
5. GitHub integration: connect a PAT, auto-create pull requests, and track draft PR status
6. Enterprise-grade: per-tenant isolation, OIDC (PKCE) authentication, AES-256 encryption, and append-only audit logs
1. Pre-development pipeline: run ECC-SDLC commands (scope → SRS → SDS → STS → estimate → proposal) with live output streamed over Server-Sent Events and artifacts stored in the cloud
2. Two-pass AI code generation: Pass 1 scaffolds the full project skeleton, Pass 2 implements each design component, powered by the Claude Agent SDK and model-agnostic (Claude, GLM, and more)
3. Browser IDE: a six-pane code-review and preview environment with file-level accept / reject
4. Safe execution: every generation runs in a resource-limited Docker sandbox with network allow-listing
5. GitHub integration: connect a PAT, auto-create pull requests, and track draft PR status
6. Enterprise-grade: per-tenant isolation, OIDC (PKCE) authentication, AES-256 encryption, and append-only audit logs
03 The result
Teams go from an idea to specced, estimated, and code-generated software inside one secure workspace, planning and building unified, AI models interchangeable, and every run reviewable and auditable. Built as three codebases: a React 19 client, a Node.js / Express bridge, and a Next.js Browser IDE, with a Python / FastAPI model gateway.