THE WORLD'S FIRST RECURSIVE CLI

Built for Bigger Terminal Jobs.

R-CLI uses recursive AI agents to decompose complex terminal work, execute autonomously, verify results, and keep going until the task is done.

R-CLI Demo
demo mode
r-clirecursive agent runtime
no commands executed
user@r-cli ~/Projects/demo % r-cli
R-CLI (v0.1.0-alpha)
model: recursive-agent-runtimedirectory: ~/Projects/demo
Prompt
ready|recursive-agent-runtime|think:max|demo mode|$0.00

Proof bar

Built for benchmark-leading terminal performance

Designed for impressive results on Terminal Bench, with third-party validation at the center of the product story.

Section: The problem

Most AI coding tools are built for short interactions

Today's AI developer tools are good at helping with the next step:

SHORT INTERACTIONS

write a function

explain an error

suggest a command

patch a small bug

BIGGER TERMINAL JOBS

They require planning, execution, verification, recovery, and persistence across multiple steps. As tasks get longer, most tools become harder to manage. Context balloons, token usage climbs, costs rise, and reliability drops.

That is where R-CLI comes in.

Section: What R-CLI is

A new kind of command line

R-CLI is not just another AI CLI.

It is a Recursive CLI: a command line powered by recursive AI agents that can decompose large goals into smaller sub-tasks, execute them in the terminal, verify results, and continue iterating until the work is complete.

GOAL INPUT

r-cli "fix failing test suite and verify release readiness"

RECURSIVE LOOP

↳ decompose goal into sub-tasks

↳ execute in terminal

↳ verify results

↳ continue iterating

Give it a goal, not just a command.

How it works

Recursive by design

R-CLI approaches terminal work as a loop:

1. Decompose

Break a large goal into manageable sub-tasks.

1. Decompose

Break a large goal into manageable sub-tasks.

2. Execute

Take action in the terminal and codebase.

2. Execute

Take action in the terminal and codebase.

3. Verify

Run checks, inspect outputs, validate progress, and identify failures.

3. Verify

Run checks, inspect outputs, validate progress, and identify failures.

4. Recurse

If the task is not complete, create new sub-tasks and continue.

4. Recurse

If the task is not complete, create new sub-tasks and continue.

This recursive execution model is what makes R-CLI different. It is not optimized for one-shot answers. It is optimized for getting complex work across the finish line.

Why recursion wins

Bigger jobs require focused context

Most AI tools carry too much context forward as tasks grow. Over time, the model is forced to work with more irrelevant state, which increases token usage, raises cost, and makes actions less reliable.

R-CLI is context-efficient by design.

Its recursive architecture keeps each sub-task focused on the context it actually needs, instead of repeatedly pulling the full workflow into the model window. That means the system stays sharper over long jobs.

Why this matters:

lower token usage

lower inference cost

less context pollution

more reliable model actions

stronger performance on long, multi-step workflows

When context stays focused, the model stays effective.

Why developers will care

Built to finish what others start

R-CLI is designed for developers who need more than command suggestions.

It is built for workflows like:
Fixing a broken test suite across multiple files

R-CLI decomposes the failing test suite into individual failures, patches affected files, re-runs checks, and recurses until the full suite passes.

Implementing a feature and validating it end to end

Give R-CLI a feature goal and it will scaffold, implement, run tests, and verify the feature works across your stack before reporting completion.

Investigating and debugging complex failures

R-CLI traces errors across logs, code, and configuration, generates hypotheses, applies fixes, and verifies each fix actually resolves the issue.

Setting up infrastructure and verifying it works

From provisioning to configuration to health checks, R-CLI handles multi-step infrastructure setup and validates each component is operational.

Handling long terminal workflows without constant supervision

R-CLI manages autonomous multi-step terminal workflows, retrying on failure and continuing through the task tree without requiring you to babysit every step.

Instead of stopping after the first output, R-CLI keeps working through the problem.

With R-CLI, you get:

stronger performance on long-horizon terminal tasks

autonomous sub-task handling

retry and recovery behavior

lower token consumption

lower cost per workflow

more reliable execution over time

Alpha invite

Join the first wave

We are opening early access to developers who want to push AI tools on real terminal workflows.

R-CLI Alpha Preview
demo mode
r-clirecursive agent runtime
no commands executed
user@r-cli ~/Projects/alpha % r-cli
R-CLI (v0.1.0-alpha)
model: recursive-agent-runtimedirectory: ~/Projects/alpha
Prompt
ready|recursive-agent-runtime|think:max|demo mode|$0.00