Etch
Etch is a signed audit chain for AI agent decisions. It records every agent action - the prompt, the context supplied, the tools called, and the resulting outcome - as a signed event. Each event is wrapped in a hybrid Ed25519 + FIPS 205 SLH-DSA-SHA2-128f signature envelope, hash-chained into epochs, and epoch-anchored to public infrastructure so that records remain verifiable long after they are written. Verification is designed to be offline. An open-source reference verifier (etch-verify) lets a third party parse audit chains and export bundles on their own laptop against pinned public keys, with no network access and no vendor cooperation required. Etch cites a 760 MB manifest verified in roughly 42 seconds with under 200 MB peak RSS using streaming ijson parsing. The product exposes three surfaces: a CLI verifier for terminals, MCP tools that drop into AI coding agents (Claude Code, Cursor, Cline, Codex, Continue, Copilot, Windsurf, Goose and others), and a web dashboard for reviewing, replaying, and sharing read-only links. Regulator-facing primitives include chain-of-custody export bundles aligned with Federal Rule of Evidence 902 and EU eIDAS, HSM/TPM/enclave attestation (tpm2, YubiKey, AWS Nitro, GCP Shielded, Azure Attestation, SEV-SNP, Intel TDX), cross-chain federation between organizations, and signed postmortems. Etch positions itself as an evidence layer that proves what happened rather than a policy engine that decides what should happen. A free tier covers 500 events over 14 days with no signup, and a persistent project can be created at any time.
- Category
- AI Agents
- Pricing
- Free
Problem solved
AI agents make consequential decisions in production, but the logs that record those decisions are easy to alter and often prove little after the fact. A plain hash chain can show that entries were reordered or tampered with, yet it cannot establish who signed what. Etch addresses this by producing signed, epoch-anchored records of agent activity that a third party - an auditor, regulator, or court - can verify offline against pinned public keys without asking the vendor for cooperation.
Use cases
- Record AI coding agent prompts, context, tool calls, and outcomes as signed events before shipping to production
- Produce signed evidence bundles for compliance reviews mapped to EU AI Act, SOC 2, and ISO article or control IDs
- Investigate and document prompt-injection attempts or anomalous agent behavior
- Give external auditors offline, read-only verification of agent history without granting platform access
- Share expiring read-only audit links with reviewers who have no account
- Federate agent events across organizational boundaries using chain_id, epoch, and event_hash references
- Attach HSM, TPM, or enclave attestation blobs to agent events in regulated environments
- Write signed postmortems and retroactive annotations into the audit chain
Audience
- Developers
- Enterprise
Pros
- Hybrid post-quantum signatures (Ed25519 + FIPS 205 SLH-DSA-SHA2-128f) with offline verification against pinned public keys
- Free tier with 500 events over 14 days and no signup required
- MCP tools integrate with a wide range of AI coding agents
- Open-source reference verifier can be run by auditors without vendor cooperation
- Compliance mapping cites specific EU AI Act, SOC 2, and ISO article and control IDs
Cons
- Early-stage product with limited traction (tens of thousands of signed events across a small number of projects)
- Provides integrity and authorship evidence only; it does not validate whether an agent decision was correct
- Open-source core repository has few stars and a small community
- Advanced capabilities such as HSM attestation, cross-chain federation, and custody bundles appear aimed at enterprise tiers
- Instrumenting existing agent pipelines requires MCP or API integration work
Official website: https://etch.systems/
Canonical LambdaBase page: https://www.lambdabase.com/tools/etch