Documentation

5-tier configuration cascade (DEFAULT to GLOBAL to USER to PROJECT to PROFILE), /use vs /profile, and inspection tools.

OpenAI-compatible endpoints with verified configurations for llama.cpp, vLLM, and SGLang.

OpenAI ChatGPT and Codex subscriptions via OAuth device code login and token refresh.

GitHub Copilot CLI subscription via OAuth device-code login without API key token metering.

Claude Code CLI subscription via OAuth session login and local token daemon.

xAI Grok Build and SuperGrok subscription via official Grok CLI device authentication.

Google Antigravity CLI subscription with Gemini and Claude Sonnet over Google Cloud Code Assist.

OpenCode Zen subscription with free contributor models via OAuth device-code login.

Direct Anthropic API and custom endpoints with four authentication modes and model aliases.

Enterprise Vertex AI publisher endpoint via Google Cloud API key with pay-per-use billing.

Scheduled callbacks, autonomous agent loops, and the /callbacks UI.

Live draggable popup UI, prompt cache warming, and background keepalive pings.

Filesystem-native sessions, deterministic compaction, and why raw sessions beat lossy summaries.

Deterministic line-based file editing tools and developer autonomy without sandbox lock-in.

Official zero data collection guarantee across SCORPIOX CODE and SCORPIO BOT operational tiers.

Zero telemetry, zero external tracking, 100% local filesystem sessions.

Complete remote API transparency with verbatim on-disk HTTP traffic recording.

Remote fleet monitoring, headless C API, and web UI.

Dynamic skill discovery, 4-tier cascade, and /skills popup.

Folder-based lifecycle event hooks with sync and async execution.

CLAUDE.md vs AGENTS.md precedence and prompt cache stability.

Native git worktree isolation per named session - own checkout, own branch, and its own session store, so parallel agents on one repository never collide and sessions stay resumable.

Terminal dashboard and headless CLI to start, watch, poke, and kill agent sessions.

Rootless Linux user-namespace isolation with no daemon, no VM, and no external dependencies.

Auto, interrupt, and queue submit modes across the TUI, the session inbox, and the embedding API.

MCP 2.0 stdio and Streamable HTTP clients, OAuth 2.1 PKCE login, and MCP server mode.

Session and thread GUIDs stamped onto OpenAI-compatible requests via OPENAI_EXTRA_HEADERS, with explicit opt-out.

One native binary as web server, app runtime, auth gate and deployment: file-layout routing, git-push deploy, JWT, MCP mode.

Live status bar, stats.json contract, embedding status events, and opt-in usage tracking that is off by default.

TOOL_SEARCHSKILLS drops the skill catalog from the prompt; SearchSkills discovers skills on demand.

/reasoning_effort <low|medium|high|max|off> for the live session plus per-provider config keys for a permanent choice.