AI platform

In development

Odin

One control plane for local models, free cloud services and paid AI providers. Odin decides where each request should run, and keeps privacy and cost in check.

Personal platform · In active development · Not publicly distributed

What it does

Explainable routing

Each request goes to the most suitable model, using a local registry, bounded benchmarks and evidence-based ranking. Odin can explain why it chose that model.

Local first

Local models are the default. Cloud providers are opt-in and disabled until configured.

Privacy gates

Requests are classified for privacy, and sensitive work can be restricted to local models or to providers that keep no data.

Budgets that hold

Spending is tracked atomically, and Odin checks free-tier pricing before every call, refusing the request if it can't confirm it. Unexpected bills don't happen quietly.

Memory for coding agents

Through an MCP server, each project gets its own durable memory of past decisions and conventions. Agents can only propose new memories, and nothing is saved until you approve it.

Context planning

Before a request runs, Odin plans its context: which memories to include, what to compress and how to stay within the model's budget.

Approach

Validated in stages

Odin is built in small, numbered stages. Each one has its own validation record, and nothing reaches the network or spends money without an explicit opt-in.

  • Fails closed: if a check can't be confirmed, the request is refused.
  • Projects are isolated, so one project's memory is never visible to another.
  • Nothing machine-specific is ever written into a repository.

Availability

Odin is a personal platform that runs Newbster Code's own development work. It isn't publicly distributed at the moment.