provedex-pipecat

provedex-pipecat is a Pipecat FrameProcessor that signs every frame in your voice agent pipeline using the Provedex sidecar. One line of integration code. Hash-chained, Ed25519-signed audit ledger as output. Built for regulated voice agents: healthcare scribes, financial voice bots, claims handlers.

The binding translates Pipecat frames into AgentEvent shapes and POSTs them over loopback HTTP to the Provedex sidecar. The sidecar holds the signing key and ledger. Your pipeline code never touches a key.

Quickstart

pip install provedex-pipecat

from provedex_pipecat import ProvedexConfig, ProvedexFrameProcessor

processor = ProvedexFrameProcessor(config=ProvedexConfig())
# Add `processor` anywhere in your Pipecat pipeline.

Assumes provedex-agent is running on 127.0.0.1:8765 (the default). To start the agent:

provedex-agent

Override the URL via the PROVEDEX_AGENT_URL environment variable or the agent_url constructor argument.

Frame mapping

Skipped frames (not signed):

• AudioRawFrame - too high frequency; hashing every audio chunk would saturate the ledger with noise.
• InterimTranscriptionFrame - not final; only committed transcripts are auditable.
• MetricsFrame - telemetry, not a decision event.
• SystemFrame subclasses - control flow, not agent output.
• LLMFullResponseStartFrame - used internally for pairing only.

Configuration reference

Latency budget

Test rig: 1000-frame burst with a 1 ms simulated agent response time (tests/test_async_smoke.py).

The producer just enqueues onto a deque; the background worker does the HTTP POST off the audio hot path. The signing round-trip never touches the frame's pass-through latency.

Failure modes

Counters are readable as attributes on the processor instance: processor.signed_total, processor.dropped_total, processor.overflow_total.

Architecture

This binding does not contain the signing primitive. The primitive is the Rust agent at https://github.com/provedex/provedex. The binding translates Pipecat frames into AgentEvent shapes per docs/spec/event-schema-v1.md and POSTs them to the agent over loopback HTTP. No key material passes through Python.

The agent signs each event with the operator's Ed25519 key and chains it via SHA-256 parent hashes into a local NDJSON ledger. Anyone with the public key can verify the ledger offline without contacting any external service.

Verifying the ledger

provedex verify

provedex verify --ledger ~/.provedex/ledger.ndjson

provedex verify --ledger /path/to/sandboxed/ledger.ndjson

provedex verify walks the chain, checks each Ed25519 signature, recomputes each SHA-256 parent hash, and exits 0 on success or 1 with a diagnostic on the first broken link.

Regulatory context

Tamper-evident audit logs are a direct requirement across several frameworks currently in force or taking effect in 2026. The EU AI Act Article 12 requires high-risk AI deployments to produce audit logs that are tamper-evident and retained for at least six months; enforcement applies from August 2, 2026. The Colorado AI Act (effective February 1, 2026) requires deployers of high-risk AI systems to maintain records sufficient to demonstrate compliance with consumer protection obligations. HIPAA's audit-control safeguard (45 CFR 164.312(b)) requires clinical voice agents to record and examine system activity, which for AI scribes means a verifiable transcript of every utterance processed. FINRA's 2026 examination priorities identify AI agent auditability as a focus area for broker-dealer supervision. A hash-chained, Ed25519-signed ledger satisfies the tamper-evident requirement across all four frameworks with a single integration point.

License: Apache-2.0

Main repo: https://github.com/provedex/provedex