Container Dietitian · Built for offline review rooms

Rules-first Dockerfile optimisation purpose-built for the air-gapped era.

CODI pairs stack-aware templates with an embedded llama.cpp module so you can analyse, rewrite, and justify Dockerfile improvements without touching the public internet. Every recommendation ships with metrics, diffs, and policy notes ready for a hackathon board review.

Primary audience: Hackathon evaluation boards assessing offline DevOps platforms.
Watch the 90-second overview pulled from the CODI brief.

Blueprint-style proof for the evaluation panel

We grounded every page in the official CODI documentation. The site follows the same blueprint aesthetic as the provided one-pager while surfacing the highlights reviewers care about.

Stack-aware templates

Deterministic rewrites

Templates map each supported stack to vetted Dockerfile structures so diffs remain predictable and auditable.

See template catalog →
Air-gap posture

Security-conscious by default

Air-gap mode blocks outbound calls, enforces adapter allowlists, and keeps data local for compliance reviews.

Review security guide →
Evidence package

Auditable metrics & reports

Every CLI run emits Markdown + HTML reports, JSON artefacts, and change rationales for scoring.

Inspect reporting stack →
Offline assist

Embedded llama.cpp

The Complete container ranks candidates and produces rationales without sending Dockerfiles to the cloud.

Dive into LLM module →

Pipeline reviewers can step through

The four-stage pipeline mirrors the notebook-style blueprint. Each stage links directly to the supporting documentation so reviewers can verify the implementation.

1 · Analyze & Detect

Parse Dockerfiles, detect stack + quality smells, and snapshot metadata for reports.

CLI Guide →

2 · Rewrite & Render

Use the rules catalog to generate deterministic candidates with inline policy notes.

Rules Guide →

3 · Report & Store

Simulate build impacts, capture metrics, and store artefacts under `runs/`.

Reporting →

4 · LLM Assist

Embedded llama.cpp ranks candidates and drafts rationales entirely offline.

LLM Module →

LLM-ready without leaving the venue

The Complete container layers llama.cpp on top of the Slim rules engine. Reviewers can see exactly how adapters are mounted, validated, and audited.

  • Local adapters mounted under /models with checksum validation.
  • LLM assist toggled via LLM_ENABLED and LLM_ENDPOINT.
  • Telemetry captured for every prompt / response cycle.
Complete container deep dive
Walk-through pulled from the evaluation LLM brief.

Slim vs Complete containers

Rules-only

Slim Container

Deterministic engine for CI pipelines and air-gapped stages. LLM disabled, outbound traffic blocked.

  • Primary function: rules catalogue execution.
  • Key use case: deterministic scoring in CI.
  • LLM support: disabled, zero network exposure.
Slim container guide →
Rules + LLM

Complete Container

Bundles llama.cpp, adapters, and ranking services for onsite LLM rationale without cloud calls.

  • Primary function: rules engine + local LLM.
  • Key use case: offline evaluations with rationales.
  • LLM support: enabled via local server.
Complete container guide →

Documentation-driven navigation

Every Markdown file in docs/ is now a first-class web page. Use the curated entry points below to jump into the area you care about.

Evaluation checklist

Installation reproducibility

Follow the exact steps in INSTALLATION.md or run the Slim container using the documented commands.

Verify install →

Run artefacts

Use REPORTING.md + site-map to trace where every HTML/JSON file lands after a `codi run`.

Review artefacts →

Deprecation trail

The new DEPRECATION_NOTICE.md shows how legacy notes map to this release to simplify scoring.

See mapping →

Blueprint reference

The visual language matches the supplied one-pager so the experience feels cohesive from pitch deck to live demo.

Open the reference image
Blueprint reference for CODI marketing site
Provided blueprint that inspired the border, grid, and hand-drawn accents.