Frameworks govern.
Tervos proves it.

The open verification layer for agentic AI. Every agent gets a cryptographic identity. Every action lands in a hash-chained log. Any slice exports as signed evidence that anyone can verify offline — no Tervos install, no network, no trust in the machine that produced it.

$ tervos start --dev
✓ Organization "Demo AI Corp" bootstrapped — 3 agents, hash-chained event log

$ tervos export-evidence --sign-as ceo -o evidence.json
✓ 23 events · chain intact · signed by did:key:z6Mk…

$ npx @tervos/verify evidence.json
✓ chain valid · signature valid — evidence verified offline▌

Your agent platform keeps an audit log.
Can anyone else trust it?

Platform logs are promises — a database the vendor agrees not to edit. Regulators, customers, and courts need proof. No framework ships cryptographically verifiable evidence of what agents did. Tervos does.

Four primitives. One chain.

Attaches to what you already run.

Tervos is not another agent framework. It counter-signs the ones you use.

// .claude/settings.json — one hook signs every tool call
"PostToolUse": [{ "hooks": [{ "command": "npx tervos hook" }] }]

$ tervos tap --from paperclip --url http://localhost:3000 --watch 30
✓ counter-signed 17 activities (chain head 8c41f0a2…)

$ tervos import --from paperclip ./my-company  # or pipe JSON into `tervos hook`
✓ Imported 5 agents — did:key identities, default-deny policies

Your entire AI org in one file.

# TERVOS.yaml
schema: tervos/v1
name: "My AI Company"

governance:
  autonomy_level: semi-autonomous
  budget: { monthly_limit_usd: 5000 }

agents:
  ceo:
    role: "CEO"
    capabilities: [strategy, delegation]
    budget_monthly_usd: 1000
    can_delegate_to: [engineer]
  engineer:
    role: "Engineer"
    reports_to: ceo
    capabilities: [coding, testing]
    budget_monthly_usd: 2000

Above MCP and A2A. Beside your framework.

Built for the regulatory reality of 2026.