👻 Ghost Console

Ready. MiniMind-trained context is used when Personal MiniMind consent and handoff are ready.

Save Gmail or Outlook client IDs in Connections, then start OAuth here.

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.

Paste client IDs (and secrets where needed) below — no terminal required. Device login works over LAN with no redirect setup; browser login needs the shown callback URL registered with the provider. Tokens stay local and are never shown.

Login status loads here.

Pasted API keys and mail app passwords. Values are never displayed — only labels. Pick a key as the chat model credential without re-typing it.

Save a Bluesky app password in Stored Keys (label it with your handle), then post or read your timeline. Sessions live only inside each request — nothing is kept.

Bluesky results appear here.

Read-only inbox fetch using a Stored Keys app password. Requires Personal MiniMind email-crawl consent. Headers + short snippets only; one-time codes and reset links are filtered. Every fetch is triaged into Act now / Today / FYI (free heuristics, zero tokens).

Inbox messages appear here.

Connector writes can require your one-time approval: each approval is bound to exactly one action (destination + payload hash) and burns on use. Reads never need approval. The audit trail is append-only and redacted.

Audit entries appear here.

Hand the browser/desktop to yourself: Ghost pauses its loop, blocks its own actions, and captures nothing while you log in manually with your own hands. Passwords and sessions never pass through Ghost. Release when done.

No takeover active.

Describe once, run on schedule or when matching mail arrives. Every firing is a run-history thread you can continue. Connector writes always park for your approval first.

Token meter, daily consumption, cost in USD, latency, and free-tier quota gauges. Free tiers price at $0.00 via the catalog; paid traffic accrues here first.

The daily re-check proposes chain rewrites here when providers advertise new models or retire pinned ones. Accepting rewrites a local config overlay — never code — and takes effect immediately.

Run a quality suite and watch the trend. Each run is recorded with pass/fail counts; history persists across restarts.

Two paths, same result. CSV: export from Chrome/Brave Settings → Passwords → Export, paste below. Direct: read this machine's Chromium login store (explicit consent, Windows-only, nothing leaves this computer). Map each entry to the vault or skip it.

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. Ghost runs its own OAuth engine — tokens are encrypted on this machine, never in a cloud.

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.

The Stealth loop's cloud brain: NVIDIA Nemotron reasoning on Nebius Token Factory, live web grounding via Tavily, and LangSmith tracing. Every panel works with zero keys configured — it just tells you what's missing and how to enable it.

Describe something that happened and Nemotron 3 Nano interprets it — the same UNDERSTAND call the Stealth loop makes.

Fresh web context for the agent — MCP-first with REST fallback, the same call the Stealth loop makes before reasoning.

What the Stealth loop actually reasoned, with the Nemotron insight and Tavily grounding that were attached to each decision.

Everything the terminal CLI can do, as buttons — same functions, same results, no terminal required. Tools that can take a while say so up front.

The full chat also lives in the draggable bubble (bottom-right). Hide it any time — this tab always brings it back. Long replies collapse to one line here.