AI at the Command Line
One command. Multiple iterations. The best answer — not the first one.
ora is a single binary that asks an AI model your question, critiques the answer, refines it, and repeats — until the answer is actually good. You set the budget. It does the work.
Give your AI agents full CLI power — any model, any strategy, structured output, background execution, and cost control. All from one command.
Up and running in 60 seconds
From download to first answer — nothing else to configure.
Install
One command. One binary. No dependencies, no Docker, no package manager.
$ curl -sSL https://oracommand.com/install.sh | shAsk anything
ora drafts, critiques, refines — and returns the best answer it can find within your budget.
$ ora -q "explain quantum tunneling"
✓ Answer v3 (confidence: 0.92)Built for the terminal. Built for agents.
Everything you need to get better answers from AI — from a single binary that works everywhere.
Multi-Provider
Claude, GPT, Gemini, Grok, Ollama, or any OpenAI-compatible API. One interface for every model.
$ ora --model claude-opus-4-5
$ ora --model gpt-4o --fallback claude-sonnet-4-20250514Autonomous Reasoning Loops
Not one answer — the best answer. ora iterates: answer, critique, refine, repeat. Three strategies: critique, debate, research.
$ ora -q "best DB for time series?" --strategy research
✓ v4 confidence: 0.91 — stoppedBudget & Cost Control
Set a budget per run. ora tracks every token, warns at 80%, and stops at your limit. Never overspend.
$ ora -q "analyze codebase" --budget 0.50
● $0.41 / $0.50 (82%) — warningBackground Jobs
Start a deep run, close your terminal, come back later. Attach, pause, resume, kill — full process control.
$ ora -q "audit the codebase" --bg
[ora-3] started in background
$ ora attach ora-3Prompt Crafting
Describe what you want in English. ora builds the optimal prompt, recommends flags, and generates ready-to-run commands.
$ ora --craft "weekly EV market report"
→ Crafted prompt + 3 command variantsMemory & Continuations
Chain runs together. Feed previous answers as context. Pick up where you left off with --continue.
$ ora -q "go deeper on point 3" --continue last
✓ Inherited context from ora-7Terminal Dashboard
Live TUI showing all processes, cost breakdown, and scheduled jobs. No browser needed — everything in your terminal.
$ ora dashboard
Processes | Cost | ScheduleAgent-Ready JSON Output
Stable JSON schema for pipelines, scripts, and AI agents. Pipe into jq, feed into other tools, automate everything.
$ ora -q "analyze" --output json --quiet | jq .answerFor every workflow
Whether you're exploring an idea or automating a pipeline — ora meets you where you are.
Ask questions in plain English
No prompt engineering needed. Just ask your question and ora handles the rest. Use --guide for interactive workflows.
# Simple question — ora iterates automatically
$ ora -q "explain what a reverse proxy does"# Interactive guide builds the command for you
$ ora --guide "how do database indexes work?"# Craft an optimized prompt from vague intent
$ ora --craft "I need to understand Kubernetes"Works everywhere you work
ora is a CLI binary — it plugs into anything that runs a command.
Terminal
Bash, Zsh, Fish, PowerShell. Any terminal on any OS. ora is a single binary — just run it.
Shell Scripts
Pipe data in, get JSON out. Use ora in bash scripts, Makefiles, and CI/CD pipelines.
Anthropic Claude
Claude Opus, Sonnet, Haiku — all supported out of the box. Auto-detected from model name.
OpenAI
GPT-4o, o3, and any OpenAI model. Same interface, same flags, same output format.
OpenClaw
Connect ora to OpenClaw workflows. Use ora as the reasoning engine behind your agents.
Ollama
Run local models with zero API keys. ora auto-detects Ollama on localhost. Fully offline.
Google Gemini
Gemini Pro, Flash — native integration. One flag to switch: --model gemini-2.0-pro
Any OpenAI-Compatible API
Groq, Together, Fireworks, self-hosted — anything with an OpenAI-compatible endpoint works with --endpoint.
Documentation
Everything you need to get the most out of ora.
Quick Start
Install ora, configure your first provider, and run your first reasoning loop in under 2 minutes.
Reasoning Strategies
Learn when to use critique, debate, or research strategies — and how to tune iteration depth and confidence thresholds.
Provider Setup
Configure Anthropic, OpenAI, Google, xAI, Ollama, or any OpenAI-compatible endpoint.
CLI Reference
Complete reference for every flag, subcommand, and configuration option.
Agent Integration
Use ora's JSON output in pipelines, scripts, and AI agent workflows. Stable schema, predictable exit codes.
Examples
Real-world examples: code review, market analysis, research synthesis, scheduled reports, and more.
Free vs Pro Coming Soon
ora is free to download and use — no limits on runs, models, or strategies. Pro adds power features for teams and automation.
| Feature | Free | Pro |
|---|---|---|
| Reasoning loops (critique, debate, research) | ✓ | ✓ |
| All providers (Claude, GPT, Gemini, Ollama...) | ✓ | ✓ |
| Background jobs & process management | ✓ | ✓ |
| Prompt crafting & guided workflows | ✓ | ✓ |
| Memory & continuations | ✓ | ✓ |
| Cost tracking & budgets | ✓ | ✓ |
| Terminal dashboard | ✓ | ✓ |
| JSON output for agents & pipelines | ✓ | ✓ |
| Scheduled runs (cron) | — | ✓ |
| Batch processing | — | ✓ |
| PDF export | — | ✓ |
| Cloud sync & sharing | — | ✓ |
| Web dashboard | — | ✓ |
| Priority support | — | ✓ |
About Us
Built by an ex-Bloomberg team
We spent years building real-time data systems at Bloomberg — where every millisecond matters and every answer has to be right. We saw how teams use AI today: paste a question into a chat window, get a mediocre first draft, manually rephrase, try again. Repeat until you give up or get lucky.
That's not how quality works. At Bloomberg, you don't ship the first draft. You iterate, critique, validate, and refine until the answer is actually correct. We built ora to bring that same discipline to AI.
The problem we're solving
One-shot answers aren't good enough
Chat UIs give you one response. Quality depends entirely on how well you prompt. Most people accept a mediocre first draft.
Manual iteration is slow and expensive
Rephrasing, re-prompting, copy-pasting context back in — you're doing the work the machine should be doing.
No cost visibility or control
API calls add up. Most tools have no budget controls — you find out what you spent when the invoice arrives.
Locked into one provider
Every AI tool picks a model for you. Switch providers? Learn a new tool. ora works with any model from any provider.
ora is the answer
One static binary that runs an autonomous reasoning loop against any model. You define the intent, the budget, and the iteration depth. ora does the rest — asks, critiques, refines, and returns the best possible answer within your constraints.
It also helps you build the right workflow in the first place. --craft turns vague intent into optimized prompts. --guide builds production-ready commands from plain English. A beginner and a power user get the same quality output.