Context

Your qbit agent knows your domain. Not a generic model's.

Your expertise, portable

You've spent months — maybe years — teaching your AI assistant your vocabulary, your workflows, your preferences. Every conversation calibrated it a little more. But that context is trapped in one provider's silo, and starting over means your agent forgets everything.

qbit.me changes that. Your accumulated AI expertise is yours — portable, attested, and served to your agent on hardware you control.

How it works

Extract

Upload a ChatGPT, Claude, or Hermes session export. qbit safely parses it, chunks it, and queues your selected qbit agent to classify each piece into three layers: domain vocabulary, workflow calibration, and behavioral preferences.

Scrub

qbit flags PII (emails, phone numbers, SSNs, credit cards, API keys) and potential trade secrets. You see every flag. You decide what stays, what gets redacted, and what gets rejected. Nothing is blindly censored; nothing ports until you say so.

Attest

You review the scrub queue. Bulk-approve the low-risk chunks in one click. Resolve each flagged one individually — approve as-is, redact the sensitive span, edit the text, or reject the whole chunk. When you're satisfied, you attest and finalize. qbit freezes the content and computes a SHA-256 PortabilityHash — your integrity guarantee.

Store

Your attested profile syncs to your selected qbit agent (your Raspberry Pi). Embeddings are computed locally with Ollama nomic-embed-text — no cloud embedding, no third-party fallback. The serving index and embeddings live on your hardware.

Serve

Your agent can now search your attested context at any time. When your request would benefit from your personal context — domain vocabulary, workflow preferences, behavioral calibration — the agent retrieves the most relevant attested chunks. Only finalized, attested context is ever served. Nothing the agent sees was not approved by you.

The trust layer

Nothing serves to your agent until you attest it. This is the differentiator vs. cloud memory products that ingest your data and serve it back opaquely. With qbit.me, every chunk that reaches your agent passed through a human review step — your review.

Agent drafts. You attest.

The ownership layer

The serving index and embeddings live on the selected hardware — your Raspberry Pi or BYOH host. Your runtime, your keys, your audit trail.

qbit Cloud retains your reviewed text for upload, review, sync, and account management. The full context does not exist only on the Pi — the cloud is the control plane, and reviewed text stays there so you can re-sync to a replacement device or manage your profiles from the dashboard.

The cold-start truth

New to AI? Your agent learns from your live conversations. Import your ChatGPT, Claude, or Hermes session history when you're ready. Native memory and attested context coexist — your agent gets smarter from every interaction, and you decide what becomes portable.

Multi-source import

qbit supports imports from the three major AI conversation platforms:

The parser auto-detects the source type from your filename, or you can select it explicitly in the upload dialog.

What this costs

Context management in the dashboard adds no dedicated endpoint price. When your agent searches your attested context via MCP, it follows the platform's standard mcp.tool.call pricing with the daily free allowance. See pricing for details.

Ready to port your expertise?

If you have a qbit agent claimed and running, open your dashboard, navigate to Context in the account menu, and upload your first export.

Get started See pricing