Tabslip Tabslip
GitHub Get started
Menu

Receipt fields #

A receipt is JSON signed by your Tabslip. show --json returns an envelope that holds the receipt and its signature, using the DSSE format. Decode the base64 payload to read the fields below. A download keeps the original signed bytes. Replace ./instance-data with the data directory Tabslip prints:

tabslip receipts --data-dir ./instance-data --limit 0 --json

Use a returned id with tabslip show <receipt-id> --data-dir ./instance-data for a terminal receipt, or add --json for the envelope. Open /r/<receipt-id> on the instance's Page URL for the paper view, signature results and details.

Signed base fields #

Nullable fields are JSON null when unknown. Zero means a reported or measured zero, never a substitute for missing data. Hashes are SHA-256 hex.

Field Meaning
type, id tabslip.receipt/v1 and locally generated rcpt_ id
instance.key_id, instance.tabslip Signing key id and Tabslip version
time.started, first_byte, finished Observed UTC timestamps; first byte can be null
time.ttfb_ms, total_ms Observed milliseconds to first byte and total duration
request.upstream, api, method, path Selected route, adapter, HTTP method and API path
request.model, stream Requested model and streaming flag, nullable
request.params Selected parameters, not messages or tool arguments
request.tag Optional validated client tag; omitted when absent
request.body_sha256, body_bytes Original client body hash and byte count before edits
request.modifications Recorded edits, such as usage injection
response.status, complete Observed HTTP status (nullable) and capture completeness
response.api_status Responses API terminal status, nullable
response.id, model, provider Provider-reported identifiers, nullable
response.finish_reason, system_fingerprint Provider-reported fields, nullable
response.error Null or {type, code}; no provider error message
response.body_sha256, body_bytes Observed response bytes hash and size
usage.source reported or unknown
usage.input_tokens, output_tokens Reported input/output; embeddings have no output count
usage.cached_input_tokens, cache_write_tokens Cache reads and total cache writes, nullable
usage.cache_write_1h_tokens One-hour subset of cache writes, nullable
usage.reasoning_tokens Reported reasoning subset of output, nullable
cost.source, amount, currency reported, estimated or unknown; decimal string and currency, nullable
cost.estimate, price_table Estimate retained alongside reported cost and attempted table version, nullable
findings Array of {code, level, text}

The request.params object keeps selected generation settings, such as temperature and token limits. It can also keep routing options and the number of tools. It excludes messages and tool arguments.

Tabslip hashes the body bytes it receives. For streams, it hashes the forwarded bytes rather than reconstructing answer text. Metadata parsing is capped at 10 MiB per JSON value or SSE event; forwarding and hashing continue when metadata cannot be decoded.

For Anthropic, signed input_tokens is uncached input. Displays add cache reads and total cache writes. If any category is unknown, the displayed input total is unknown. One-hour writes are already inside total writes. Signed JSON and decoded JSONL keep the raw categories used for pricing. response.complete describes capture, not successful model output: fully received Responses incomplete and failed events can be complete captures.

JSON example #

This made-up OpenRouter example shows decoded JSON before supplemental cost arrives. It has no signature. A stored receipt has full ids and hashes where this example uses shortened values:

{
  "type": "tabslip.receipt/v1",
  "id": "rcpt_…",
  "instance": {"key_id": "ts_…", "tabslip": "dev"},
  "time": {"started": "2026-10-21T09:00:00.000Z", "first_byte": "2026-10-21T09:00:00.030Z", "finished": "2026-10-21T09:00:00.050Z", "ttfb_ms": 30, "total_ms": 50},
  "request": {"upstream": "openrouter", "api": "openai.chat", "method": "POST", "path": "/v1/chat/completions", "model": "openai/gpt-4.1", "stream": false, "params": {}, "tag": "example", "body_sha256": "…", "body_bytes": 74, "modifications": []},
  "response": {"status": 200, "complete": true, "api_status": null, "id": "gen-example-unpriced", "model": "openai/gpt-4.1", "provider": null, "finish_reason": "stop", "system_fingerprint": null, "error": null, "body_sha256": "…", "body_bytes": 300},
  "usage": {"source": "reported", "input_tokens": 12, "output_tokens": 4, "cached_input_tokens": 0, "cache_write_tokens": null, "cache_write_1h_tokens": null, "reasoning_tokens": null},
  "cost": {"source": "unknown", "amount": null, "currency": null, "estimate": null, "price_table": null},
  "findings": [{"code": "price_unknown", "level": "observed", "text": "The cost is unknown."}]
}

Findings #

Code Level Meaning
model_alias_resolved reported A dated version of the requested model; informational, does not activate the findings filter
model_differs reported Requested and reported model differ
fallback_used reported Provider reports a later fallback model or multiple attempts
usage_missing observed No reported usage
price_unknown observed No reported cost or usable estimate
stream_interrupted observed Stream lacks its terminal event or client disconnected
upstream_error observed HTTP error or provider error event/object
transport_error observed Connection, TLS, read or early cancellation failure
metadata_unavailable observed Metadata could not be parsed within the limit

Expected unknown cost on demo, Ollama and custom upstreams does not itself produce price_unknown.

These describe reported values and observed failures; none checks the model itself.

Signed supplements and displayed totals #

OpenRouter can add a second DSSE envelope with payload type application/vnd.tabslip.supplement+json. Its type is tabslip.supplement/v1; receipt_id and receipt_sha256 bind it to the base payload. source is openrouter.generation and fetched is its time. Nullable provider, cost, data_region, and native_tokens.prompt and completion come from generation details. currency accompanies cost. attempts lists reported provider/status pairs; findings carries additional findings. Missing lists are empty. Provider details can arrive after the call.

The signed base receipt never changes. The page and unsigned exports count supplement cost once. They keep the existing provider when the supplement's provider is null.

When a supplement supplies a missing cost, price_unknown is marked resolved. The original finding stays signed and visible. The receipt page lists findings from the base receipt first, then from supplements. Each finding names its signed source. Findings from different sources stay separate even when their codes match. Resolved findings stay out of the Calls badge and findings filter.

The receipt page and the terminal show command verify the original receipt and every attached supplement before showing SIGNED. An invalid receipt or supplement produces INVALID with the failed check; show exits 10. show --json still returns only the original receipt envelope. If the receipt or any attached supplement fails verification, it prints diagnostics on stderr and exits 10, even when the base envelope alone is valid. Stored data that cannot be decoded is a read error (exit 2). A missing receipt uses exit 5. See verification, costs and the command reference.

Edit this page on GitHub ↗