ora Documentation

ora is a command-line tool. You give it a question and a set of flags that control how it reasons — model, strategy, budget, iterations. It does the rest.

Installation

One command. No dependencies, no runtime, no package manager.

curl -sSL https://oracommand.com/install.sh | sh

Or download the binary directly from the download page.

Verify it works:

ora --version

First Queries

ora is a CLI. You pass your question with -q and control behavior with flags. Here are real queries you can run right now:

Basic question

Just ask something. ora runs a reasoning loop with defaults (critique strategy, up to 10 iterations, $1 budget).

ora -q "what is the best database for time series data?"

Pick a model

Use --model to choose any model from any provider.

ora -q "explain quantum tunneling simply" --model gpt-4o

Control depth

Set how many iterations ora runs. More iterations = more refinement = better answer.

ora -q "design a rate limiter for a REST API" --min-iter 3 --max-iter 15

Set a budget

Cap spending per run. ora stops gracefully and returns the best answer found.

ora -q "full competitive analysis of EV market" --budget 0.50

Choose a strategy

Three strategies: critique (default), debate (two perspectives), research (breaks into sub-questions).

ora -q "postgres vs mongodb for IoT?" --strategy debate

Save the output

ora -q "summarize Q1 metrics" --save report.md

Quiet mode — just the answer

ora -q "what is a monad?" --quiet

Dry run — see what would happen

Estimate cost and time without making any API calls.

ora -q "analyze this codebase" --model claude-opus-4-5 --max-iter 10 --dry-run

Combine flags

Flags compose freely. This runs a deep research query with Claude, capped at $0.50, saving to a file:

ora -q "comprehensive analysis of AI regulation in the EU" \
  --model claude-opus-4-5 \
  --strategy research \
  --min-iter 5 --max-iter 20 \
  --budget 0.50 \
  --save eu-ai-regulation.md

Provider Setup

ora works with any AI provider. Set your API key:

# Anthropic (Claude)
ora config set anthropic-key sk-ant-...

# OpenAI (GPT)
ora config set openai-key sk-...

# Google (Gemini)
ora config set google-key AIza-...

# Or use environment variables
export ORA_ANTHROPIC_KEY=sk-ant-...

ora auto-detects the provider from the model name: claude-* → Anthropic, gpt-* → OpenAI, gemini-* → Google.

For Ollama (local models), no API key needed:

ora -q "prompt" --model llama3.2

For any OpenAI-compatible endpoint:

ora -q "prompt" --model mixtral-8x7b --endpoint https://api.together.xyz/v1

How ora Works

Instead of a single API call, ora runs an iterative loop:

iter 1:  answer(prompt)                        → answer_v1
iter 2:  critique(answer_v1) → refine(critique) → answer_v2
iter 3:  critique(answer_v2) → refine(critique) → answer_v3
...
stop:    confidence >= threshold OR budget reached OR max iterations

Each iteration, the model scores its own confidence (0.00–1.00). ora returns the highest-confidence answer across all iterations — not necessarily the last one.

Strategies

Critique (default)

Answer → critique → refine → repeat. Best for most queries.

ora -q "prompt" --strategy critique

Debate

Two perspectives (advocate + skeptic) debate, then synthesize. Best for nuanced topics.

ora -q "postgres vs mongodb for IoT?" --strategy debate

Research

Decomposes into sub-questions, answers each, synthesizes. Best for complex research.

ora -q "full EV market analysis" --strategy research

Confidence & Stopping

ora stops when any of these conditions are met:

Confidence reaches the threshold (default: 0.85)
Budget is exhausted
Max iterations reached
Convergence detected (confidence barely changing for 3 iterations)

ora always runs at least --min-iter iterations regardless of confidence.

ora -q "prompt" --confidence 0.95  # higher bar
ora -q "prompt" --min-iter 5       # at least 5 iterations

Budget Control

Set a per-run spending cap in USD:

ora -q "prompt" --budget 0.50

# Check your spending
ora cost
ora cost --by-model

ora warns at 80% of budget and stops gracefully at the limit, returning the best answer found so far.

Background Jobs

Run long queries in the background:

ora -q "deep analysis" --bg
# → [ora-3] started in background

ora list              # see all processes
ora attach ora-3      # stream output
ora status ora-3      # check progress
ora pause ora-3       # pause mid-run
ora resume ora-3      # continue
ora kill ora-3        # terminate (saves best answer)

Memory & Continue

Chain runs together using previous answers as context:

# Inject a previous run's output
ora -q "go deeper on point 3" --memory ora-1

# Inject last 3 runs
ora -q "what patterns do you see?" --memory last:3

# Continue — inherits model, system prompt, strategy
ora -q "expand on the risk section" --continue last

Prompt Crafting

Let ora optimize your prompt before running:

# Craft from intent
ora --craft "analyze the EV market every Monday"

# Improve a weak prompt
ora --craft --improve "tell me about stocks"

# Interactive workflow builder
ora --guide

--craft generates an optimized prompt plus ready-to-run commands. --guide walks you through building a workflow step by step.

Context Files & URLs

ora -q "summarize this" --context report.pdf
ora -q "critique this" --context https://example.com/article
ora -q "compare these" --context doc1.pdf --context doc2.pdf

ora strips HTML, enforces size limits, and warns about potential prompt injection.

Dashboard

Live terminal UI showing all processes, costs, and scheduled jobs:

ora dashboard

Navigate with arrow keys. Enter to attach, K to kill, P to pause, Q to quit.

CLI Reference

# Query
ora -q "prompt"                          # basic query
ora -q "prompt" --model claude-opus-4-5   # specific model
ora -q "prompt" --budget 0.50            # cost cap
ora -q "prompt" --strategy research      # strategy
ora -q "prompt" --bg                     # background
ora -q "prompt" --quiet                  # final answer only
ora -q "prompt" --output json            # JSON output
ora -q "prompt" --save report.md         # save to file
ora -q "prompt" --dry-run                # estimate only

# Process management
ora list [--all] [--running]
ora attach <id>
ora status <id>
ora kill <id>
ora pause <id>
ora resume <id>

# Tools
ora cost [--by-model]
ora history [--search "keyword"]
ora prompts [--save "name=content"]
ora dashboard
ora config show
ora config set <key> <value>

# Utilities
ora --craft "intent"
ora --guide
ora test [--dry-run]
ora export <id> --format md|json
ora --version

JSON Output

Use --output json --quiet for structured output in pipelines:

{
  "schema_version": 1,
  "id": "ora-3",
  "status": "done",
  "stopped_by": "confidence_threshold",
  "answer": "...",
  "confidence": 0.91,
  "iterations": { "completed": 4, "min": 2, "max": 10 },
  "cost": { "total_usd": 0.18, "tokens_in": 8200, "tokens_out": 4100 }
}

# Extract answer
ora -q "prompt" --output json --quiet | jq -r '.answer'

# Check confidence
ora -q "prompt" --output json --quiet | jq '.confidence > 0.85'

Configuration

Config lives at ~/.ora/config.toml:

ora config show                          # view all settings
ora config set model claude-opus-4-5      # default model
ora config set budget 1.00               # default budget
ora config set min-iter 2                # default min iterations
ora config set max-iter 10               # default max iterations
ora config set strategy critique         # default strategy
ora config set confidence 0.85           # default threshold

Priority: explicit flags > environment variables > config file > defaults.

Cost Tracking

ora tracks the cost of every run automatically. Costs are based on the token pricing set by each AI provider — ora uses whatever the model charges. You pay your provider directly; ora just tracks it for you.

# See your spending
ora cost                   # today, week, month, lifetime
ora cost --by-model        # breakdown by model

# Estimate before running
ora -q "prompt" --dry-run

# Set a budget cap
ora -q "prompt" --budget 0.50

Local models (Ollama) are free. For cloud models, check your provider's pricing page for current rates. ora will never spend more than your --budget.