← Back to tools

oracle

Bundle code and prompts for AI analysis with multi-model support

Homepage Repository
Development Productivity linuxmacoswindows TypeScript MIT

Description

Oracle packages source code files and natural language prompts into structured bundles for submission to AI models. It supports API and browser-based execution across OpenAI, Google, Anthropic, and OpenRouter models. Features include multi-model comparison runs, token budgeting, session management, async handling for long-running queries, and MCP server integration.

Install

npmnpm install -g @steipete/oracle
homebrewbrew install steipete/tap/oracle
npxnpx -y @steipete/oracle

AI Summary

Bundles code files with prompts and sends them to AI models (GPT-5, Gemini, Claude) for analysis, code review, or documentation. Supports API and browser modes, multi-model comparisons, and session persistence.

Capabilities

  • + Bundle source files with prompts for AI-assisted code analysis
  • + Multi-model comparison: run the same query across GPT-5, Gemini, Claude, and others
  • + API mode with automatic provider detection from environment variables
  • + Browser mode using Chrome cookies from signed-in ChatGPT or Gemini sessions
  • + Token budgeting with per-file usage reports
  • + Session management: save, replay, and resume interrupted queries
  • + Async handling for long-running GPT-5 Pro queries with reattach support
  • + Dry-run mode to preview bundles without sending
  • + Copy assembled bundle to clipboard for manual paste workflows
  • + MCP server mode via oracle-mcp for integration with AI coding assistants

Use When

  • → Getting AI-powered code review or analysis with full file context
  • → Comparing how different AI models answer the same question about your code
  • → Sending large codebases to AI models with proper token budgeting
  • → Working without API keys by using browser-based AI sessions
  • → Persisting AI query sessions for later review or continuation

Avoid When

  • x Simple questions that do not require file context
  • x Real-time interactive coding assistance (use an IDE copilot instead)
  • x Processing sensitive code that should not be sent to cloud AI providers

Usage Patterns

Review code with API

oracle -p "Review this code for bugs" --file src/app.ts

Sends the file with the prompt to the default AI model for review

Multi-model comparison

oracle -p "Analyze architecture" --models gpt-5.1-pro,gemini-3-pro --file "src/**/*.ts"

Runs the same query against multiple models and compares responses

Dry run to preview

oracle --dry-run summary -p "Explain this" --file src/

Previews the bundle summary without sending it to any provider

Copy bundle to clipboard

oracle --render --copy -p "Analyze this" --file docs/

Assembles the bundle and copies it to clipboard for manual pasting

Check session status

oracle status --hours 72

Lists all oracle sessions from the last 72 hours

Input / Output

stdin: Not typically used
stdout: AI model responses, session summaries, bundle previews
stderr: Progress and status messages
Exit codes:
0 Success
1 Error

Typical Pipelines

oracle -p "Review for security issues" --file "src/**/*.ts" --json | jq .response
oracle --render --copy -p "Document this" --file src/main.ts

Related Tools

summarize complementary

Summarize handles URL/media content; Oracle handles code file bundles
View AGENTS.md for oracle