oracle 🧿 — Whispering your tokens to the silicon sage
Oracle bundles your prompt and files so another AI can answer with real context. It speaks GPT-5.1 Pro (default alias to GPT-5.2 Pro on the API), GPT-5.1 Codex (API-only), GPT-5.1, GPT-5.2, Gemini 3 Pro, Claude Sonnet 4.5, Claude Opus 4.1, and more—and it can ask one or multiple models in a single run. Browser automation is available; use --browser-model-strategy current to keep the active ChatGPT model (or ignore to skip the picker). API remains the most reliable path, and --copy is an easy manual fallback.
Quick start
Install globally: npm install -g @steipete/oracle
Homebrew: brew install steipete/tap/oracle
Requires Node 22+. Or use npx -y @steipete/oracle … (or pnpx).
# Copy the bundle and paste into ChatGPT npx -y @steipete/oracle --render --copy -p "Review the TS data layer for schema drift" --file "src/**/*.ts,*/*.test.ts" # Minimal API run (expects OPENAI_API_KEY in your env) npx -y @steipete/oracle -p "Write a concise architecture note for the storage adapters" --file src/storage/README.md # Multi-model API run npx -y @steipete/oracle -p "Cross-check the data layer assumptions" --models gpt-5.1-pro,gemini-3-pro --file "src/**/*.ts" # Preview without spending tokens npx -y @steipete/oracle --dry-run summary -p "Check release notes" --file docs/release-notes.md # Browser run (no API key, will open ChatGPT) npx -y @steipete/oracle --engine browser -p "Walk through the UI smoke test" --file "src/**/*.ts" # Gemini browser mode (no API key; uses Chrome cookies from gemini.google.com) npx -y @steipete/oracle --engine browser --model gemini-3-pro --prompt "a cute robot holding a banana" --generate-image out.jpg --aspect 1:1 # Sessions (list and replay) npx -y @steipete/oracle status --hours 72 npx -y @steipete/oracle session <id> --render npx -y @steipete/oracle restart <id> # TUI (interactive, only for humans) npx -y @steipete/oracle tui
Engine auto-picks API when OPENAI_API_KEY is set, otherwise browser; browser is stable on macOS and works on Linux and Windows. On Linux pass --browser-chrome-path/--browser-cookie-path if detection fails; on Windows prefer --browser-manual-login or inline cookies if decryption is blocked.
Integration
CLI
- API mode expects API keys in your environment:
OPENAI_API_KEY(GPT-5.x),GEMINI_API_KEY(Gemini 3 Pro),ANTHROPIC_API_KEY(Claude Sonnet 4.5 / Opus 4.1). - Gemini browser mode uses Chrome cookies instead of an API key—just be logged into
gemini.google.comin Chrome (no Python/venv required). - If your Gemini account can’t access “Pro”, Oracle auto-falls back to a supported model for web runs (and logs the fallback in verbose mode).
- Prefer API mode or
--copy+ manual paste; browser automation is experimental. - Browser support: stable on macOS; works on Linux (add
--browser-chrome-path/--browser-cookie-pathwhen needed) and Windows (manual-login or inline cookies recommended when app-bound cookies block decryption). - Remote browser service:
oracle serveon a signed-in host; clients use--remote-host/--remote-token. - AGENTS.md/CLAUDE.md:
- Oracle bundles a prompt plus the right files so another AI (GPT 5 Pro + more) can answer. Use when stuck/bugs/reviewing. - Run `npx -y @steipete/oracle --help` once per session before first use. - Tip: set
browser.chatgptUrlin config (or--chatgpt-url) to a dedicated ChatGPT project folder so browser runs don’t clutter your main history.
Codex skill
- Copy the bundled skill from this repo to your Codex skills folder:
mkdir -p ~/.codex/skillscp -R skills/oracle ~/.codex/skills/oracle
- Then reference it in your
AGENTS.md/CLAUDE.mdso Codex loads it.
MCP
- Run the stdio server via
oracle-mcp. - Configure clients via steipete/mcporter or
.mcp.json; see docs/mcp.md for connection examples.
npx -y @steipete/oracle oracle-mcp
- Cursor setup (MCP): drop a
.cursor/mcp.jsonlike below, then pick “oracle” in Cursor’s MCP sources. See https://cursor.com/docs/context/mcp for UI steps.
{
"oracle": {
"command": "oracle-mcp",
"args": []
}
}Highlights
- Bundle once, reuse anywhere (API or experimental browser).
- Multi-model API runs with aggregated cost/usage, including OpenRouter IDs alongside first-party models.
- Render/copy bundles for manual paste into ChatGPT when automation is blocked.
- GPT‑5 Pro API runs detach by default; reattach via
oracle session <id>/oracle statusor block with--wait. - Azure endpoints supported via
--azure-endpoint/--azure-deployment/--azure-api-versionorAZURE_OPENAI_*envs. - File safety: globs/excludes, size guards,
--files-report. - Sessions you can replay (
oracle status,oracle session <id> --render). - Session logs and bundles live in
~/.oracle/sessions(override withORACLE_HOME_DIR).
Browser auto-reattach (long Pro runs)
When browser runs time out (common with long GPT‑5.x Pro responses), Oracle can keep polling the existing ChatGPT tab and capture the final answer without manual oracle session <id> commands.
Enable auto-reattach by setting a non-zero interval:
--browser-auto-reattach-delay— wait before the first retry (e.g.30s)--browser-auto-reattach-interval— how often to retry (e.g.2m)--browser-auto-reattach-timeout— per-attempt budget (default2m)
oracle --engine browser \ --browser-timeout 6m \ --browser-auto-reattach-delay 30s \ --browser-auto-reattach-interval 2m \ --browser-auto-reattach-timeout 2m \ -p "Run the long UI audit" --file "src/**/*.ts"
Flags you’ll actually use
| Flag | Purpose |
|---|---|
-p, --prompt <text> |
Required prompt. |
-f, --file <paths...> |
Attach files/dirs (globs + ! excludes). |
-e, --engine <api|browser> |
Choose API or browser (browser is experimental). |
-m, --model <name> |
Built-ins (gpt-5.1-pro default, gpt-5-pro, gpt-5.1, gpt-5.1-codex, gpt-5.2, gpt-5.2-instant, gpt-5.2-pro, gemini-3-pro, claude-4.5-sonnet, claude-4.1-opus) plus any OpenRouter id (e.g., minimax/minimax-m2, openai/gpt-4o-mini). |
--models <list> |
Comma-separated API models (mix built-ins and OpenRouter ids) for multi-model runs. |
--base-url <url> |
Point API runs at LiteLLM/Azure/OpenRouter/etc. |
--chatgpt-url <url> |
Target a ChatGPT workspace/folder (browser). |
--browser-model-strategy <select|current|ignore> |
Control ChatGPT model selection in browser mode (current keeps the active model; ignore skips the picker). |
--browser-manual-login |
Skip cookie copy; reuse a persistent automation profile and wait for manual ChatGPT login. |
--browser-thinking-time <light|standard|extended|heavy> |
Set ChatGPT thinking-time intensity (browser; Thinking/Pro models only). |
--browser-port <port> |
Pin the Chrome DevTools port (WSL/Windows firewall helper). |
| `--browser-inline-cookies[(-file)] <payload | path>` |
--browser-timeout, --browser-input-timeout |
Control overall/browser input timeouts (supports h/m/s/ms). |
--browser-recheck-delay, --browser-recheck-timeout |
Delayed recheck for long Pro runs: wait then retry capture after timeout (supports h/m/s/ms). |
--browser-reuse-wait |
Wait for a shared Chrome profile before launching (parallel browser runs). |
--browser-profile-lock-timeout |
Wait for the shared manual-login profile lock before sending (serializes parallel runs). |
--render, --copy |
Print and/or copy the assembled markdown bundle. |
--wait |
Block for background API runs (e.g., GPT‑5.1 Pro) instead of detaching. |
--timeout <seconds|auto> |
Overall API deadline (auto = 60m for pro, 120s otherwise). |
--background, --no-background |
Force Responses API background mode (create + retrieve) for API runs. |
--http-timeout <ms|s|m|h> |
HTTP client timeout (default 20m). |
--zombie-timeout <ms|s|m|h> |
Override stale-session cutoff used by oracle status. |
--zombie-last-activity |
Use last log activity to detect stale sessions. |
--write-output <path> |
Save only the final answer (multi-model adds .<model>). |
--files-report |
Print per-file token usage. |
--dry-run [summary|json|full] |
Preview without sending. |
--remote-host, --remote-token |
Use a remote oracle serve host (browser). |
--remote-chrome <host:port> |
Attach to an existing remote Chrome session (browser). |
--youtube <url> |
YouTube video URL to analyze (Gemini browser mode). |
--generate-image <file> |
Generate image and save to file (Gemini browser mode). |
--edit-image <file> |
Edit existing image with --output (Gemini browser mode). |
--azure-endpoint, --azure-deployment, --azure-api-version |
Target Azure OpenAI endpoints (picks Azure client automatically). |
Configuration
Put defaults in ~/.oracle/config.json (JSON5). Example:
{ model: "gpt-5.1-pro", engine: "api", filesReport: true, browser: { chatgptUrl: "https://chatgpt.com/g/g-p-691edc9fec088191b553a35093da1ea8-oracle/project" } }
Use browser.chatgptUrl (or the legacy alias browser.url) to target a specific ChatGPT workspace/folder for browser automation.
See docs/configuration.md for precedence and full schema.
Advanced flags
| Area | Flags |
|---|---|
| Browser | --browser-manual-login, --browser-thinking-time, --browser-timeout, --browser-input-timeout, --browser-recheck-delay, --browser-recheck-timeout, --browser-reuse-wait, --browser-profile-lock-timeout, --browser-auto-reattach-delay, --browser-auto-reattach-interval, --browser-auto-reattach-timeout, --browser-cookie-wait, --browser-inline-cookies[(-file)], --browser-attachments, --browser-inline-files, --browser-bundle-files, --browser-keep-browser, --browser-headless, --browser-hide-window, --browser-no-cookie-sync, --browser-allow-cookie-errors, --browser-chrome-path, --browser-cookie-path, --chatgpt-url |
| Run control | --background, --no-background, --http-timeout, --zombie-timeout, --zombie-last-activity |
| Azure/OpenAI | --azure-endpoint, --azure-deployment, --azure-api-version, --base-url |
Remote browser example
# Host (signed-in Chrome): launch serve oracle serve --host 0.0.0.0:9473 --token secret123 # Client: target that host oracle --engine browser --remote-host 192.168.1.10:9473 --remote-token secret123 -p "Run the UI smoke" --file "src/**/*.ts" # If cookies can’t sync, pass them inline (JSON/base64) oracle --engine browser --browser-inline-cookies-file ~/.oracle/cookies.json -p "Run the UI smoke" --file "src/**/*.ts"
Session management
# Prune stored sessions (default path ~/.oracle/sessions; override ORACLE_HOME_DIR)
oracle status --clear --hours 168More docs
- Bridge (Windows host → Linux client): docs/bridge.md
- Browser mode & forks: docs/browser-mode.md (includes
oracle serveremote service), docs/chromium-forks.md, docs/linux.md - MCP: docs/mcp.md
- OpenAI/Azure/OpenRouter endpoints: docs/openai-endpoints.md, docs/openrouter.md
- Manual smokes: docs/manual-tests.md
- Testing: docs/testing.md
If you’re looking for an even more powerful context-management tool, check out https://repoprompt.com
Name inspired by: https://ampcode.com/news/oracle
