TruthLayer

Evidentiary telemetry
for autonomous AI agents.

Cryptographically chained. Signed by keys the agent never holds. Anchored to Bitcoin. Independently verifiable without our permission.

14,638

Sustained events per second. 99.98% success.

Zero silent failures. Production-verified 22 July 2026.

Regulated entity Partner Investor

Whitepaper  ·  Verify our claims yourself (ledger-verify on GitHub)

The APRA standard for AI evidence.

On 30 April 2026, the Australian Prudential Regulation Authority issued a letter to regulated entities calling for “continuous, auditable evidence of AI control rather than point-in-time, sample-based assurance.”

The tools most entities currently use cannot meet that standard.

AI observability

Buyer
Engineering
Evidence
SDK-instrumented traces
Cryptographic guarantees
None
Retention
Days to weeks

Reference vendors: Langfuse, Arize, Datadog LLM Obs

AI governance suites

Buyer
Compliance
Evidence
Questionnaires and integrations
Runtime enforcement
On roadmap
Retention
Depends on source

Reference vendors: Credo AI, Holistic AI, OneTrust

Traditional GRC

Buyer
Internal audit
Evidence
Integration pulls and attestations
Runtime AI evidence
None
Retention
Attestation records only

Reference vendors: Vanta, Drata, ServiceNow

Continuous evidence must be produced by the runtime itself, and must be tamper-evident, jurisdiction-aware, and independently verifiable.

The network is the witness.

TruthLayer captures agent actions at the Cloudflare edge — AI Gateway, Monetization Gateway, MCP-over-Workers, Sandbox SDK, Logpush — before they reach any downstream system. The agent needs no SDK, holds no signing key, and cannot suppress the record.

TruthLayer architecture Agent traffic flows through Cloudflare edge products into the ingest worker, then a single per-agent Durable Object with SQLite which holds the hash chain, certified by a per-tenant Durable Object. Artefacts land in R2 — private payloads, a public registry, and OpenTimestamps proofs anchored to Bitcoin. The ledger-verify tool runs offline against the public artefacts alone. Agent AI Gateway Monetization Gateway MCP over Workers Sandbox SDK Logpush ingest-worker AgentDO SQLite · hash chain signs with subkey TenantDO certifies subkeys R2 private payloads public registry · seals OTS proofs Bitcoin block ledger-verify runs offline
Capture at the edge; one Durable Object call per event; artefacts verifiable offline.

Five design principles, from the whitepaper.

  1. 1 Substrate-native, not SDK-based.
  2. 2 One event, one chain slot, always.
  3. 3 Per-tenant cryptographic isolation.
  4. 4 Publicly verifiable from artefacts alone.
  5. 5 Kill switches actually cut.

Measured, not projected.

Every performance claim on this page is measured from a 10-minute distributed load test conducted 22 July 2026 against the live production deployment — 9.51 million requests generated from 16+ Cloudflare locations worldwide.

Load-test results across three architecture generations
4-hop architecture Single-hop (single machine) Distributed edge fleet
Sustained req/s 4,728 6,042 14,638
Success rate 99.82% 99.99% 99.98%
p50 latency ~180ms 25–59ms 143ms
CPU per event 1–4ms (server) not re-measured 1–2ms, flat
Silent failures 0 0 0

p99 at 14,638 req/s with worldwide traffic is 1.8s — attributable to Durable Object data placement relative to suddenly-global callers, not compute. The whitepaper states this in full; so do we.

Verify our claims yourself.

The ledger-verify CLI has zero Cloudflare dependencies and reconstructs and verifies the entire chain from public artefacts alone.

View on GitHub → Read the whitepaper →

Choose your path.

Regulated entities

For Heads of Operational Risk at APRA-regulated institutions running or planning autonomous AI agents in production.

12-week structured pilot programme. A$19,999 fixed. CPS 230 evidentiary package delivered week 12. Three design-partner slots open for Q4 2026.

Request pilot invitation

Partners

For Big-4 AI assurance practices, AI liability insurance MGAs, and Cloudflare partnerships teams building customer trust for autonomous AI systems.

Revenue share, evidence-source agreements, or co-selling — depending on your business model.

Discuss a partnership

Investors

For AI infrastructure investors interested in the runtime-evidence category and the insurance-underwriting flywheel that reprices autonomous-agent liability.

Investor materials sent on request. Not currently raising a priced round.

Request investor materials