# decionis.ai

> Open-source runtimes, conformance specifications, SDKs and verification tools built on the Decionis execution-authority protocol.

decionis.ai is the developer side of Decionis. The commercial site (evaluation, deployment, trust, pricing) is decionis.com, and the protocol and API reference is the Decionis docs.

## Projects

- AgentSafe (Apache-2.0, release v0.3.5, runtime 0.2.5): execution gateway for AI agents and APIs. Source: https://github.com/decionis/agent-safe-pipeline
- Steward (Apache-2.0, v0.3.0): governed customer-operations review surface. Source: https://github.com/decionis/steward. Demo: https://decionis-steward.vercel.app
- Verify (@decionis/verify 0.3.0 on npm, Apache-2.0; source not public): offline verification for Decision Dossiers and proof artifacts. Package: https://www.npmjs.com/package/@decionis/verify
- Agent-Safe Intent (agent-safe.intent/1): open execution-intent conformance specification, rendered on this site from https://github.com/decionis/agent-safe-pipeline at v0.3.5 (b4d5c9e).

## Pages

- [decionis.ai](https://decionis.ai): Install AgentSafe, test the boundary locally, and find each project.
- [AgentSafe](https://decionis.ai/agentsafe): The execution gateway: install, the outcome of each request, shadow and enforcement, the library, examples.
- [Steward](https://decionis.ai/steward): The customer-operations review surface: what stays with Decionis, how to run it.
- [Verify](https://decionis.ai/verify): Check a Decision Dossier's signatures offline, and what a pass does and does not mean.
- [Specifications](https://decionis.ai/specs): Specifications published with their test vectors.
- [Agent-Safe Intent](https://decionis.ai/specs/agent-safe-intent): The execution-intent format, its versions and how it changes.
- [Agent-Safe Intent v1](https://decionis.ai/specs/agent-safe-intent/v1): The specification text, rendered from the AgentSafe repository.
- [Agent-Safe Intent v1: changes](https://decionis.ai/specs/agent-safe-intent/v1/changelog): Every change to v1, with its pull request and vector.
- [Conformance](https://decionis.ai/conformance): The conformance corpus and what OpenAI, Vercel AI SDK, LangChain and MCP tool calls carry of a binding.
- [SDKs](https://decionis.ai/sdks): Every Decionis SDK and package, with its license and whether its source is public.
- [Adapters](https://decionis.ai/adapters): Framework tool calls, the MCP guard, the EffectAdapter contract and provider-side verification.
- [MCP execution interception](https://decionis.ai/adapters/mcp): How each MCP tools/call invocation is bound to an intent and runs once, on a grant.

## Try it

- Install: brew tap decionis/agent-safe https://github.com/decionis/agent-safe-pipeline && brew trust decionis/agent-safe && brew install agentsafe
- Test the boundary locally, no account: agentsafe test
- Check the Agent-Safe Intent corpus offline: agentsafe verify intent conformance/vectors conformance/agent-safe-intent-v1.json
- Verify a Decision Dossier offline: npx @decionis/verify --file dossier.json --jwks https://api.decionis.com/.well-known/decision-dossier-jwks.json

## Elsewhere

- Commercial site: https://decionis.com
- Protocol and API reference: https://decionis.com/docs/api
- Failure posture by surface: https://decionis.com/trust/failure-posture

Facts checked against each repository and registry on 2026-09-25.