Bandicot

Fifty tool calls, written as one script.

When a job is a loop over many files or results, Bandicot's agent writes one short script that calls its tools, filters what comes back, and returns only the answer.

  • One script instead of fifty tool calls
  • Approval before writes
  • See what every turn spent
  • Fewer tokens for the same work· In development

Solo is free for one profile · providers bill you directly

Sign upLog inPricing

An agent that checks forty files makes forty calls, reads forty results into its context, and carries every one of them into the next turn.

The script does the loop. The conversation keeps the answer, not the raw results, and every call inside still goes through your approvals.

A big, repetitive job takes one step, and the chat keeps the answer instead of the raw results.

Why it holds
The script runs in a sandbox with no files, network or processes of its own; it reaches only the agent's tools.
How soon
Reads inside one script run in parallel, and nothing waits on the model between calls.
What it saves you
No watching the agent page through results one call at a time.

What does the work, and what is still being built.

Every card names its status. Available now means you can use it today; in development and direction mean it is not shipped yet.

Free Bandicot core

Available now

Solo is free with every capability for one profile. Bring a provider key, a local model, or both; a provider's own bill stays yours.

Paid plans add profiles to an account. Bandicot does not resell model tokens.

A trusted harness around the model

Available now

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

The native host owns execution facts. The model proposes work; it does not certify a write, spend, or success.

Projects that hold the work

Available now

Keep files, context, plans, and decisions with the project instead of losing them in a disposable chat.

A session is anchored to a project boundary so a pivot starts from an intentional context, not accidental history.

Durable work sessions

Available now

Close the window, restart, or pick up from another of your Macs without turning the work into a brand-new conversation.

The host records session state and every window reattaches to it. When a long chat is summarized, the agent can still search and reread what was there.

A human gate for writes

Available now

When work can change files or systems, Bandicot makes the decision visible before the effect lands.

The approval boundary fails closed when an action is not allowed. The model cannot grant itself permission.

Scripts for repetitive tool work

Available now

For a loop over many files or results, the agent writes one short script that calls its tools and returns only the answer.

The script runs in a sandbox with no files, network or processes of its own. Every tool it calls goes through the same approvals.

Agents with a project boundary

Available now

Let an agent inspect, plan, and use tools inside the project you selected instead of an unbounded chat transcript.

The host controls tool calls, cancellation, permissions, and the final recorded result.

Plans before tool work

Available now

Turn a vague request into an inspectable sequence of steps before tools begin changing the project.

The harness tracks dependencies and blocked steps. A conversation alone does not silently advance execution.

Every token in view

Available now

See the tokens, cache reads and cost of every turn, what each session cost and was worth, and how much each account has left.

Bandicot reports what the provider or local runtime exposes. A provider that hides usage shows a dash, never a guess.

For you if

Developers who ask agents to check, search or change many things at once.

Not for you if

One quick question: a single tool call is simpler, and the agent makes one.

Script Runner

Solo · free · 1 profile

  • A program tool in every session
  • A sandbox with no files, network or processes
  • The same approvals for every call inside
  • Every call a script made, on the record

Free for one profile, no card required. Add profiles when someone joins.

Solo is free for one profile · providers bill you directly

Sign upLog inPricing

Focus is a feature.

Bandicot does these too, but they are not why this page exists. Anything marked in development or direction is not shipped yet.

  • Shared compute, intentionally routed
  • Optional access controls
  • Nested profiles
  • Teams across machines

Questions before you start

Where does the script run?

On your Mac, in a small JavaScript sandbox inside the engine. It has no filesystem, network, process or timer of its own, and reaches only the tools the session already has.

Can a script do something I did not approve?

No. Each tool it calls goes through the same approvals as a direct call, and file changes still land as versions you can undo.

When does the agent use it?

For several related calls, loops over collections, or filtering large results. For one call, or steps that need judgment in between, it calls the tool directly.

Does it save tokens?

It keeps raw results out of the conversation, and each result reports how much the script read next to what it returned. We do not print a whole-task saving until our fixed-corpus measurement is published.

The script does the loop. The conversation keeps the answer, not the raw results, and every call inside still goes through your approvals.

Solo is free for one profile · providers bill you directly

Sign upLog inPricing

Fifty tool calls, written as one script.: Bandicot for Developers who hand agents jobs over many files, logs or results.
Fifty tool calls, written as one script.. Every claim on this page names its status.

Solo is free for one profile · providers bill you directly

Start free