bandicot

Bring the server. Skip the agent shell.

Configure your OpenAI-compatible endpoint, pick its model for a session, and the host supplies the project, the tool approvals and the file history. You keep the model side; you do not write the rest.

Free for one person on your Mac, no card. Your model provider bills you directly.

What you get
Your endpoint with a project workspace around it.
Why it holds
The host owns the tools and approvals, and hosted providers and local models can share one project.
How soon
Configure the endpoint and pick its model for a session.
How much work
No writing a new agent shell around your inference server.
CustomOpenAI-compatibleendpoint supportTool approvals ownedby the hostDurable projectsessionsEndpoint Workspaceasks firstoutside your rules

Before and after

Now

You run your own inference server and then spend weeks building the agent loop, the permissions and the history around it.

With Bandicot

Model choice stays apart from the tools. Your endpoint answers, the host runs the tools under your approval rules, and prompts go to the address you configured and nowhere else.

Endpoint Workspace

Every line below is already in the plan. Nothing here is an extra you pay for.

  • Custom OpenAI-compatible endpoint supportincluded
  • Tool approvals owned by the hostincluded
  • Durable project sessionsincluded
  • Saved git history brought back as it wasincluded
  • Free for one profileincluded
Free1 profileFree

Free for one person on your Mac, no card. Your model provider bills you directly.

No contract. Cancel any time from your account and keep the plan until the end of the period you paid for.

The whole idea in one line

Hosted providers and local models, in the same project.
The host owns the tools and approvals, and hosted providers and local models can share one project.

Read this before you start

Made for

Developers who run and assess their own endpoint.

Developers connecting self-managed OpenAI-compatible model servers

Not yet for

People seeking hosted inference or universal protocol compatibility.

One working day, start to finish

No writing a new agent shell around your inference server.

  1. Early

    The job lands

    You run your own inference server and then spend weeks building the agent loop, the permissions and the history around it.

  2. Before noon

    You hand it over

    Model choice stays apart from the tools. Your endpoint answers, the host runs the tools under your approval rules, and prompts go to the address you configured and nowhere else.

  3. Mid-afternoon

    A trusted harness around the model

    Bandicot gives every agent a project, tools, a session that survives restarts, approvals, and a clear place to stop.

  4. Last thing

    You read what changed

    Your endpoint with a project workspace around it.

Bring the server. Skip the agent shell.

Free for one person on your Mac, no card. Your model provider bills you directly.