TTTPS: deterministic evidence receipts through pre-ingestion cryptographic sealing

Copypip install litellm-openttt

Copyimport litellm
from litellm_openttt import TTTPSTimestampLoggerKPPLive

litellm.callbacks = [TTTPSTimestampLoggerKPPLive()]

resp = litellm.completion(
    model="gpt-3.5-turbo",
    messages=[{"role": "user", "content": "hi"}],
)
print(resp.tttps_receipt)

Copypip install llamaindex-openttt

Copyfrom llama_index.core.callbacks import CallbackManager
from llamaindex_openttt import TTTPSTimestampCallbackHandlerKPPLive

Settings.callback_manager = CallbackManager([TTTPSTimestampCallbackHandlerKPPLive()])

resp = llm.complete("hi")
print(resp.additional_kwargs["tttps_receipt"])

Not a Python import: ollama-openttt is a reverse proxy that sits in front of ollama serve and seals the final NDJSON line of a stream. Point any Ollama client at the proxy port instead of the daemon port.

Copypip install ollama-openttt

Copyexport KPP_API_KEY=...

ollama serve &              # upstream, port 11434
ollama-openttt --port 11435  # or: python -m ollama_openttt

curl http://127.0.0.1:11435/api/chat -d '{
  "model": "qwen2.5:0.5b",
  "messages": [{"role": "user", "content": "hi"}],
  "stream": false
}'

ASGI middleware — wraps any Starlette/FastAPI app in front of a vLLM server and attaches a receipt to each request span.

Copypip install vllm-openttt

Copyfrom starlette.applications import Starlette
from vllm_openttt import TTTPSMiddleware

app = Starlette(routes=your_routes)
app.add_middleware(TTTPSMiddleware)

A DeepEval metric — seals the test case and scores whether its receipt has valid TTTPS provenance structure.

Copypip install deepeval-openttt

Copyfrom deepeval.test_case import LLMTestCase
from deepeval_openttt import TTTPSProvenanceMetric

metric = TTTPSProvenanceMetric()
score = metric.measure(your_test_case)

A Ragas metric — same provenance-scoring shape as the DeepEval adapter, for teams already standardized on Ragas evaluation.

Copypip install ragas-openttt

Copyfrom ragas import evaluate
from ragas_openttt import TTTPSProvenanceMetric

results = evaluate(your_dataset, metrics=[TTTPSProvenanceMetric()])

A standard OpenTelemetry SpanProcessor — AutoGen's runtime tracing is built on OTel already, so this attaches to any TracerProvider AutoGen (or any other OTel-instrumented Python stack) is already using, no framework-specific adapter needed. On every span end it extracts the trace/span id, model name, and a BLAKE3 hash of the input/output (never the raw text itself) and seals that into the Proof-of-Time chain in a background thread — export is never blocked.

Copypip install ttt-otel

Copyfrom opentelemetry.sdk.trace import TracerProvider
from ttt_otel import TTTSpanProcessor

provider = TracerProvider()
provider.add_span_processor(
    TTTSpanProcessor(api_key="<YOUR_TTT_API_KEY>")  # or os.environ["TTT_API_KEY"]
)

It runs alongside your existing OTLP exporter — you don't replace your observability backend, you add a second processor next to it. Confirmed compatible today:[MEASURED: pypi.org/pypi/ttt-otel/json, project description field, 2026-08-16]

Verifying a seal later — if a span's output was edited after sealing, this returns valid: false:

Copyfrom ttt_otel import verify
from ttt_otel._hash import canonical_fields_from_span

result = verify(canonical_fields_from_span(span), api_key="<YOUR_TTT_API_KEY>")
print(result["valid"], result["chain_intact"])