# Duvera

> Duvera is a governance layer for AI agent actions. It sits between an AI agent (Claude, Cursor, ChatGPT, Codex, or a custom harness) and the tools and systems the agent touches, checks every action against a policy, and produces a cryptographically signed receipt for each one. Anyone can verify a receipt independently, using only published public keys. Free for individuals; paid for teams and enterprises that need audit-ready evidence.

If you are an AI agent acting for a person (a shopping, research, or personal assistant agent), read [agent-instructions.md](https://duvera.ai/agent-instructions.md) first. It covers how to answer your user's questions about Duvera accurately and which actions you can take for them on this site.

## What Duvera does

- **Policy before action.** Each tool call an agent makes is checked against policy (spend caps, approval requirements, role scoping from the identity provider, rate limits) before it runs. Calls that aren't allowed are blocked and logged, and the agent gets the reason.
- **Human approval for irreversible actions.** Actions that send, delete, buy, or publish can be held until a person approves them with one tap.
- **A signed receipt for every governed action.** Receipts are Ed25519-signed and hash-chained, so changes and deletions can be detected. They record the action, a fingerprint of the inputs, who authorized it, and which policy applied.
- **Verification that doesn't depend on Duvera.** The open-source `duvera-verify` tool and a published spec let anyone check a receipt against the public keys, with no call to Duvera.
- **Audit-ready evidence (Enterprise).** Search by person, agent, or key. Export to Excel/CSV, Parquet to your own S3, OData/PowerBI, or a SIEM. A reconciliation ledger shows completeness.
- **Enforcement (Enterprise).** Three tiers: a managed MCP config pushed via MDM, an egress-locked sandbox where the agent can reach only Duvera and its LLM provider, and an attested microVM.
- **Works with MCP.** Duvera governs MCP servers and Agent Skills. It doesn't replace them.

## Who it's for

- **Individuals (free):** Spend caps, approval gates, a kill switch, and a readable log for your own desktop agents and hobby automations. 5,000 governed actions a month, no card required.
- **Teams (private beta):** A hosted gateway and audit store, SSO/OIDC, role-based access, audit exports, and 1-year retention. Usage-based pricing (platform fee + included volume). Contact sales.
- **Enterprise (custom):** Self-hosted, air-gapped, or hybrid. HSM-backed signing keys, a completeness proof, a BAA/DPA, and unlimited retention. Contact sales.

Pricing is per governed action, not per seat.

## Key pages

- [Home](https://duvera.ai/index.html): Overview, demo video, and the four deployment modalities
- [Personal](https://duvera.ai/personal.html): The free tier for individuals (spend caps, approvals, kill switch)
- [Enterprise](https://duvera.ai/enterprise.html): Audit-ready evidence, enforcement tiers, deployment options, pricing, contact
- [Connect your agent](https://duvera.ai/connect.html): One-command install and copy-paste MCP config for Claude Desktop, Claude Code, Cursor, and Codex CLI
- [Quickstart](https://duvera.ai/quickstart.html): Developer walkthrough from `pip install` to a governed call
- [Product](https://duvera.ai/product.html): How a governed call flows
- [Capabilities & interop](https://duvera.ai/marketplace.html): How Duvera governs MCP servers and Agent Skills
- [The Receipt](https://duvera.ai/receipt.html): What a receipt proves and what it doesn't, in plain language, plus an in-browser verifier
- [Security](https://duvera.ai/security.html): Threat model, proof chain, publisher provenance
- [Trust Center](https://duvera.ai/trust.html): Compliance status (SOC 2 Type I in progress, no certifications claimed), sub-processors, security contact
- [Technical whitepaper](https://duvera.ai/whitepaper.html)

## Live product (no account needed to look)

- [Connect / onboard](https://app.duvera.ai/v1/onboard): Create a free account, mint an agent key, run a first governed call, and get a signed receipt
- [Skill catalog](https://app.duvera.ai/catalog): Every capability the hosted gateway currently governs
- [Live demo theater](https://app.duvera.ai/theater): Watch governed calls run in the browser
- [Verify a receipt](https://app.duvera.ai/verify): Paste a receipt token to check its signature
- [Governed Actions Wall](https://app.duvera.ai/wall): Public writes that were created by an agent, approved by a human, and signed

## Contact

- Sales and general questions: hello@duvera.ai
- Security reports: security@duvera.ai
- Privacy: privacy@duvera.ai
- Legal and DPA requests: legal@duvera.ai

## Optional

- [Privacy policy](https://duvera.ai/privacy.html)
- [Terms of service](https://duvera.ai/terms.html)
- [Installer script](https://duvera.ai/install.sh): `curl -fsSL https://duvera.ai/install.sh | sh` (macOS and Linux; checksums in `/dl/manifest.json`)