👻 Ghost Console

Create a session to let Ghost join meetings or interviews with disclosure, consent, and Trust Runtime journaling.

Config is local to this machine. Raw secret values are never displayed here.

Connect GitHub for repo-aware planning, verified skill discovery, and self-evolution source review.

GitHub OAuth opens a browser/device approval flow. Tokens stay local and are never shown.

Use read-only email OAuth for Personal MiniMind. Save client IDs here, then connect Gmail or Outlook.

Email crawling requires MiniMind email-crawl consent and a completed provider login.

Choose a path to synthesize Ghost Chimera.

Remote commands are paired, audited, and policy-gated. Provider adapters are optional. Signing secrets make provider webhooks fail closed.

Trust Runtime is local-first. Refresh to inspect run journals, approvals, MCP trust, evals, and trace export readiness.

GitHub sign-in is optional. Existing GHOSTCHIMERA_GITHUB_TOKEN, GITHUB_TOKEN, or gh CLI auth still works.

Preview mode only: Ghost will not ingest, install, train, or activate material without approval.

Connect GitHub to plan issue-to-PR work.

Host execution is off by default. Arm it only for trusted local work you are willing to audit and revert.

Click a shortcut to load it into the objective box, then hit Run.

Ready.

Paste a raw RFC 2822 email below, or provide a path to a .eml / .mbox file on the server. Ghost learns from your email context automatically.

Provide a path to a text file (.txt, .md, .py, .json, …) or a directory to recursively ingest. Ghost chunks and indexes the content for future context.

Ready.

Ghost's bi-temporal knowledge graph holds durable facts promoted from episodic memory. Run a sleep-time consolidation pass to promote salient memories and retire stale facts.

Record a good prompt/response pair to grow Ghost's personal training dataset. This data can later be used to fine-tune a local MiniMind model on your domain.

Paste a JSON array of {"instruction"|"prompt", "output"|"response"} records and export as a JSONL training file for local MiniMind fine-tuning workflows.

Whole machine crawl scans supported readable files under crawl roots using normal OS permissions, default exclusions, and file limits.

Train a local neural personal adapter from approved MiniMind dataset records. This updates local adapter weights; it does not upload data or mutate upstream model weights.

Gmail and Outlook use read-only OAuth scopes. Tokens stay local and are never shown in API responses.

Ready.

Ready.

Log in with Slack, Notion, GitHub, Gmail and more. Connect natively with OAuth or 1-click via Nango — Ghost polls connected services into background events.

Select Refresh Status to load the provider catalog.

Ready.

Skills are specialist capabilities Ghost can use. Click Run Skill to invoke one directly, or just mention the skill name in a Run objective.

Search GitHub for skill candidates, then click Convert on a candidate to scaffold a compatible Ghost skill.

Ready.

Preview only: Ghost will not download, convert, train, or activate a model without approval.

Operational trace only. Hidden chain-of-thought is never exposed.

Built-in tools run locally and do not require external MCP servers.

Runs a local preview journey and preserves findings.

Ready.

New schedules start disabled — enable them after reviewing the cron expression.

Run these commands locally before tagging or pushing a beta release.

Live feed from the background event loop: events observed, decisions made, interventions prepared, outcomes learned. Silence is a successful outcome.

Every draft is rewritten in plain human language (ASD-STE100 core rules) so it never reads as AI-generated. Edit freely, copy-paste anywhere, or approve to send.

Simplified text and applied rules appear here.