Why this matters
When engineers and reviewers ask LLMs about code or design, useful context is often scattered across files, tests, and notes — pasting fragments leads to confusion, missing references, and noisy answers. Oracle addresses that friction by assembling a deterministic, token‑aware bundle of the exact files and prompt you care about, then delivering that bundle to one or more models (API or automated browser) so answers are grounded in the real project context.
What Sets It Apart
- Multi‑model first: run the same bundled context against combinations like gpt-5.5-pro, gpt-5.2-pro, gemini-3.1-pro, and Claude Sonnet in a single command, producing per‑model outputs and an aggregate oracle.json with success/failure metadata — useful for comparative audits.
- API + browser automation: API mode uses provider keys (OpenAI, Gemini, Anthropic) for reliability; browser mode automates Chrome to use ChatGPT/Gemini UI without requiring API keys, capturing transcripts and artifacts when needed.
- File-aware bundling and safety: supports globs, excludes, size guards, token usage reports, and bundles as text or ZIP so the receiving model actually sees the intended context rather than ad hoc pastes.
- Session lineage and tooling for reproducibility: sessions are stored under ~/.oracle/sessions with transcripts, artifacts, perf traces, and followup chaining (followup from response ids or browser sessions) so audits and iterative reviews are reproducible.
Who It's For — Fit & Tradeoffs
Great fit if you are an engineer, reviewer, or small team who: wants to run repeatable, context-rich prompts over real project files; needs side‑by‑side outputs from multiple modern models; or wants to automate one‑off code/architecture audits without building custom tooling. Oracle is CLI-native (Node.js/TypeScript), integrates with MCP/Codex workflows, and includes developer conveniences like --files-report, --render --copy, and --followup chaining.
Look elsewhere if you need a turnkey web UI for non‑technical users, a hosted enterprise governance layer out of the box, or a turnkey production inference server — Oracle is a developer tool that assumes CLI familiarity, Node 24+ runtime, and (for stable API runs) valid provider keys. Browser automation is powerful but marked experimental on non‑macOS platforms; file uploads and cookies may require extra platform flags on Linux/Windows.
Practical notes
- Platform & prerequisites: Node 24+; browser automation is most stable on macOS; API runs require OPENAI_API_KEY / GEMINI_API_KEY / ANTHROPIC_API_KEY as appropriate.
- Typical workflows: one‑line bundle+copy for manual ChatGPT paste; multi‑model API panels for cross‑model audits;
oracle status+oracle session <id>for replay and lineage. - Privacy & ops: sessions and artifacts are saved locally under ~/.oracle/sessions by default; review storage and browser cookie choices before running sensitive workloads.
In short: Oracle reduces the friction of giving models real project context, makes multi‑model comparisons and reproducible consults straightforward, and trades a polished end‑user GUI for a powerful, scriptable CLI workflow geared toward engineers and reviewers.
