A small, low-cost AWS platform you install into your account.
Point an AI coding assistant at the stack repo and it does the rest.

A lightweight AWS stack plus a handful of drop-in features that all live behind one dashboard URL. Clone the stack repo, open it in an AI coding assistant, and about half an hour later you have:

• a dashboard on CloudFront with one card per installed feature
• a shared API-key store so every feature authenticates the same way
• an admin console for creating users, issuing scoped AWS access keys, and managing per-user context

Most of that wall clock is CloudFront waiting for eventual consistency on each new distribution — there's nothing to actively babysit.

Idle cost with all features installed is about $1/month — Lambdas, CloudFront, DynamoDB on-demand, and a handful of small S3 buckets. No VPC, no NAT gateway, no load balancer in the default mode.

1. Get an AWS account and run aws configure. That's the only thing you need before starting.
2. Clone tokenburner-ai/stack and open it in an AI coding assistant (one that reads CLAUDE.md — such as Claude Code or any other assistant that supports project-level context files).
3. Say "install tokenburner." The assistant reads CLAUDE.md, checks your prerequisites, asks which features you want, bootstraps CDK if needed, and runs the real CLI (python3 tokenburner.py install). You approve the AWS deploys as they happen.
4. Open the dashboard URL the install prints. Every feature card uses the same admin API key, which is cached at ~/.tokenburner/credentials.

If you prefer the command line, the same CLI runs without an assistant — no prompts, no .tokenburner.json to write, just:

git clone https://github.com/tokenburner-ai/stack.git && cd stack
pip install pyyaml --break-system-packages   # macOS Homebrew Python
python3 tokenburner.py install               # uses your AWS CLI defaults
# or pin the target explicitly:
python3 tokenburner.py install --profile myprofile --region us-east-2

The CLI auto-bootstraps CDK if your target region isn't ready, runs a Bedrock pre-flight before deploying chat, and prints a dashboard URL plus admin API key when it finishes.

                           ┌──────────────────────────────────────┐
User → CloudFront (HTTPS) →│  Dashboard Lambda (one card/feature) │
                           └──────────────────────────────────────┘
                                          │
                                          ▼
                           ┌──────────────────────────────────────┐
                           │ Feature Registry DDB   API Keys DDB  │
                           └──────────────────────────────────────┘
                                          ▲
                 ┌────────────────────────┼────────────────────────┐
                 │                        │                        │
         Drive Lambda             Chat Lambda              Agent Lambda
         + CF + S3 + DDB          + CF + Bedrock + DDB    + CF + DDB
                 │                        │                        │
                 └──── self-register via custom resource ──────────┘

Each feature is an independent CDK stack. It imports the shared exports from the base, then writes one row to the feature-registry table on deploy. The dashboard reads that table and renders one card per row — that's the entire plugin model.

• Install, then forget. Dev mode runs on Lambda + CloudFront + DDB on-demand. Nothing costs money when idle.
• AI-assistable by construction. Every repo has a CLAUDE.md or tokenburner.md that tells an AI coding assistant exactly what it's looking at and how to change it. Works with any assistant that reads project-level context — the stack doesn't depend on a specific vendor.
• Features are repos. Adding one means copying drive as a template, renaming, and appending an entry to features.yaml.
• Brand-neutral code. AI vendor names appear only in onboarding prose. The code, the API, and the Administration console name nothing specific — you can point any backend at the Agent feature.

tokenburner.ai