"What was that charge?"
Precise, structured-record answers. Sub-second. Routes to the typed-records query path. Returns the record + a verifiable seal.
Platform · Who else reads the ledger
We don't want to be the chatbot. We want to be the layer every chatbot trusts.
Lossless speaks MCP and REST. Every read is gated by per-scope consent (read receipts but not messages; read vehicles but not finance) and per-tool rate limits. Claude, GPT, Gemini, and Apple Intelligence can all read from the same sealed ledger — the user decides who, what, and for how long. Every read is written to the audit log.
← Any agent that speaks MCP / REST
Consent is scoped, revocable, audited
Same sealed ledger everyone reads →
How it's different
Other AI chats guess. Ours cites.
An "AI assistant" without your records is a creative-writing partner. It will make up plausible-sounding numbers because it has nothing real to draw from. Every Lossless response — whether served back to the user's own chat or to a third-party agent over MCP — is grounded in typed records. Every answer is reconstructable from cited sources.
If a question can't be answered from the data, the API returns an explicit empty result. No confidently wrong numbers.
- ✓ Every answer ties to one or more typed records
- ✓ Citations link to the original email, PDF, or bank entry
- ✓ "I don't have data on this" is a valid response — and a frequent one
- ✓ Hallucinated-citation rate: 0.6% (vs. 4.1% on flat retrieval)
Worked examples
Real questions. Sourced answers.
Each of these can be asked by the user in Lossless chat, or by a third-party agent over the Agent API — the response payload is the same, the citation chain is the same, the seal is the same.
What did I pay for tires last year?
Answer · 4 sources
You spent $1,847.30 on tires across 2 vehicles in 2025: $401 (Costco, May), $389 (Costco, Nov, Y), $612 (Discount Tire, Aug, 3), and $445 (Tesla service, Mar, 3 alignment).
When was the Y last serviced?
Answer · 3 sources
Apr 12, 2026. 12V battery replaced under warranty at Tesla Service · Fremont. Cost $0.00. Odometer at service: 38,402 mi. Open recall (front camera firmware) was closed during the same visit.
How much did I spend at superchargers in March?
Answer · 14 sessions
$284.41 across 14 sessions, 312 kWh delivered. 8 sessions on Model Y, 6 on Model 3. Most-used station: Stockton (5×, $96.40).
Show me everything for my CPA.
Answer · export ready
I assembled 219 line items for tax year 2025: vehicle expenses ($4,210), home office ($1,820), charitable contributions ($2,400), 1099-DIV ($840), 1099-B (87 trades). Each line cites its source PDF + email + bank entry.
Modes
Three shapes of question. One API.
"What did I spend on…"
Sums, groupings, time windows. The model writes the query; we audit it before running. Returns rows + the SQL + the consent scope used.
"Tell me about…"
Prose explanations grounded in retrieved records. Always cites. Never fabricates. Response payload includes the source-record IDs so any client can re-render the citations.
"Per-scope consent + per-tool rate limits + an audit log on every read is the security-review checkbox you've been asking us for. Now we just integrate it like any other OAuth provider."
— Technical evaluator · multi-tenant agent platform
Continue the architecture tour
You've seen the surface. Now go back to the map.
That's the architecture tour. The Provenance Ledger is the center of all of it; every pillar feeds in or reads out. Head back to the Overview to see the whole shape in one diagram.