OPC is a CLI product for people who want to run a small AI product team from a terminal. You talk to Olivia, your executive assistant. Olivia turns your plain-language request into coordinated work across product, business, risk, interface, design, implementation, testing, review, and final delivery.

The goal is simple: you should be able to speak like a boss, not like a prompt engineer.

OPC is powered by the OpenHands SDK, uses a lightweight OpenSpec-style planning layer, and can connect to Stitch for interface exploration when configured.

What OPC is for

Use OPC when you want an AI team to build or improve a small software product, prototype, internal tool, landing page, or purpose-built business agent.

OPC is designed for tasks such as:

• Turning a customer problem into a scoped product prototype.
• Building a local software tool from a business description.
• Creating a polished homepage or interface concept.
• Revising an existing project after the first delivery is not good enough.
• Keeping product, engineering, testing, and review roles separated.
• Producing a boss-facing delivery report instead of an internal agent log.

How it works

OPC is organized as a small company with named employees:

• Olivia receives your request and protects your attention.
• Victoria checks whether the work makes business sense.
• Nora turns the request into a product spec.
• Maya handles interface direction and Stitch-assisted design thinking.
• Leo looks for risk, edge cases, and operational problems.
• Sophie validates customer success and delivery expectations.
• Ada designs the technical approach.
• Miles turns the spec into an implementation task plan.
• Devin writes the code.
• Tess tests the result.
• Quinn reviews the diff and decides whether changes are mergeable.
• Iris writes the final boss-facing delivery report.

The team runs as an actor-style workflow. Each employee has a role, prompt profile, task inbox, and boundaries. Employees pass work to each other through task handoffs instead of all sharing one uncontrolled conversation.

Quick Start

Install OPC globally:

npm install -g @katetech-99/opc

OPC uses uv to run its Python workspace:

uv --version

If uv is missing, install it from the official uv documentation:

curl -LsSf https://astral.sh/uv/install.sh | sh

Configure your LLM and optional Stitch credentials:

opc setup

Start OPC from the project directory you want the team to work on:

cd /path/to/customer-project
opc

Or pass a workspace explicitly:

opc /path/to/customer-project

Then speak to Olivia naturally:

We have a local coffee brand that wants a high-end homepage for subscription
beans. Make it feel premium, warm, and conversion-focused. I only need a local
prototype, but it should look like something I can show the client.

Configuration

The setup wizard writes:

~/.opc/config.env

Required values:

OPENAI_BASE_URL
OPENAI_API_KEY

Optional Stitch values:

STITCH_API_KEY
STITCH_ACCESS_TOKEN
GOOGLE_CLOUD_PROJECT

Environment variables override values in ~/.opc/config.env.

Useful setup commands:

opc setup
opc doctor
opc config
opc config path
opc --version

Interactive commands

Inside the OPC CLI:

/help                 Show commands.
/workflow             Show the current company workflow.
/employees            Show employees and prompt files.
/edit <name>          Edit an employee's soul prompt.
/edit <name> system   Edit an employee's system suffix prompt.
/project new <name>   Create a project context.
/project open <slug>  Restore a previous project context.
/project status       Show the current project context.
/repo <path>          Change target workspace.
/status               Show current workspace and project state.
/exit                 Leave OPC.

While the team is working:

/interrupt <request>  Stop the current run and restart with new direction.
/revise               Enter a multi-line revision.
/status               Show which employees are working or waiting.
/cancel               Stop the current run.

Workspaces and artifacts

OPC keeps installation files, user workspaces, and run artifacts separate.

By default, npm-installed OPC stores run records and isolated workspaces here:

~/.opc/artifacts/my_team

You can override this location:

export OPC_ARTIFACTS_DIR=/path/to/opc-artifacts

When possible, OPC creates a git worktree for each task. If the target workspace is not a safe git worktree, OPC creates an isolated snapshot with a local git baseline. This avoids treating the npm installation directory, a parent home directory repo, or a cache directory as the product workspace.

Delivery style

OPC's final output is written for the boss:

• What was actually built.
• Whether the result satisfies the customer request.
• How to run or inspect the product.
• What risks remain.
• What should happen next.

Internal logs, event JSON, and run directories are not treated as product deliverables.

Development

This package is built from a uv-managed Python workspace with a Node launcher.

Run tests:

uv run pytest tests/my_team

Run lint and type checks for changed files:

uv run pre-commit run --files <paths>

Preview the npm package:

npm pack --dry-run

License

MIT