llms.txt
Compare with other sites' llms.txt files →# Keydris
> Keydris is authorization infrastructure for AI agents. It lets developers control what AI agents can do before they do it.
Keydris is authorization infrastructure for AI agents. You define the authority, Keydris checks it before the action runs, and the action runs in the system that owns it. The mechanism is action-time authorization: every governed agent action presents a KIT (Keydris Identity Token), checked against your policy for scope and expiry before it executes. Authority is evaluated at the authorization boundary: a broker you run on your side, a KIT Reader on the receiving side, or a Keydris integration where the Reader cannot be installed. No path completes that decision without the Keydris platform today: the enforcement point requests verification, and the control plane makes the authority decision against current policy and revocation state. Each check returns one of three decisions: ALLOW, the governed action proceeds; REJECT, the governed action does not proceed; APPROVAL REQUIRED, held for a human and resolving to ALLOW or REJECT. Every decision is recorded as an audit entry. ALLOW means the authority boundary permitted the attempt; it does not mean the underlying tool succeeded, and the record keeps the decision and the execution outcome as separate facts. When a decision cannot be obtained, the current platform fails closed: the action does not proceed; that platform-unavailable behavior is currently fixed and is separate from a policy's own configured default decision.
You define the policy that sets an agent's authority; Keydris issues a KIT bound to that policy and evaluates it at the authorization boundary. That evaluation depends on the Keydris platform today: no path completes the authority decision without it, and where the decision cannot be reached the boundary fails closed.
On the broker and Reader paths Keydris is not a proxy in front of your traffic: your payloads stay inside your boundary, and only KIT verification, the policy check, and revocation state reach the Keydris platform. Where the Reader cannot be installed on a third-party system such as GitHub or Slack, Keydris provides a managed integration that sends the final provider API request using the organization's connected credential, which is securely stored and never reaches the agent. On those managed paths Keydris is in the data path, so Keydris is not universally out of it.
## What Keydris is not
Every statement below is a boundary, not a disclaimer. Each one exists because
the opposite is a reasonable guess that happens to be wrong.
- Keydris is not an identity provider. It binds authority to an identity your
systems already issue, and it depends on OAuth, IAM and workload identity
rather than replacing them.
- Keydris is not a secrets manager or a credential broker. Where an integration
needs a provider credential, the agent does not receive it, but this is a
per-integration architecture and not a universal claim that Keydris never
handles credentials.
- Keydris does not execute agent actions. It returns the authorization
decision; the action runs in the system that owns it.
- Keydris is not a universal proxy and does not see all traffic. It governs
actions that use a configured Keydris authority path. A request arriving
without Keydris authority may still use the receiving system's normal
authentication, so "every action is governed" is not true.
- Keydris is not an agent runtime, an agent framework, or a monitoring product.
- Keydris is not universally out of the data path: on managed integrations such
as GitHub and Slack it participates in the execution path.
- ALLOW is an authorization decision, not an execution result. An allowed action
can still fail downstream, and the record keeps those as separate facts.
## Getting started
Install with `npm install --global @keydris/cli`. Setup runs `keydris login`,
then `keydris init`, then `keydris proxy`. The quickstart walks through Claude
Code and OpenAI Codex; the reference page carries the full command surface.
These pages are authoritative, and this file routes to them rather than
restating them.
- [Quickstart](https://keydris.com/docs/quickstart): install the CLI, connect a registered agent, start the proxy, and run a first governed session
- [CLI and vocabulary reference](https://keydris.com/docs/reference): installation and updates, setup and identity, proxy and session commands, inspection and cleanup, the files the CLI changes, and the exact meaning of each term
- [@keydris/cli on npm](https://www.npmjs.com/package/@keydris/cli): the public CLI package
## Concepts
- [Introduction](https://keydris.com/docs/introduction): the authority model, and why an authenticated agent can still do the wrong thing
- [KITs](https://keydris.com/docs/concepts/kit): what a Keydris Identity Token binds (identity, scope, expiry)
- [Policies](https://keydris.com/docs/concepts/policies): how owner-defined, versioned policies decide each action
- [Execution boundary](https://keydris.com/docs/concepts/execution-boundary): where a governed action stops until authority is verified, and why the boundary rests closed
- [Audit logs](https://keydris.com/docs/concepts/audit): the record of every decision
- [KIT management](https://keydris.com/docs/concepts/authority): issuing and revoking KITs
## Integrations
- [MCP](https://keydris.com/docs/integrations/mcp): governing MCP tool calls and resource reads under the same identity, policy and evidence model
- [GitHub](https://keydris.com/docs/integrations/github): governing what agents may do in selected repositories
- [Slack](https://keydris.com/docs/integrations/slack): governing what agents may read and post in selected public channels
- [Local commands](https://keydris.com/docs/integrations/local-commands): supported shell commands checked before the agent's harness executes them, returning ALLOW, REJECT or APPROVAL REQUIRED
## Operations
- [Console](https://keydris.com/docs/operations/console): where operators manage agents, policies, integrations, sessions, evidence, members and billing
## Site
- [AI agent authorization](https://keydris.com/agent-authorization): the canonical explanation of the category. What agent authorization is, why identity is not authority, what a governed action is, what ALLOW / REJECT / APPROVAL REQUIRED decide, what Keydris does and does not do, and how it relates to OAuth, IAM and MCP
- [Home](https://keydris.com/): what Keydris is, and how a governed action moves from identity through agents and policy to a decision (ALLOW, APPROVAL REQUIRED or REJECT)
- [Contact](https://keydris.com/contact): reach the team about what you are building, or book a meeting to go through architecture or security requirements before a production decision
- [FAQ](https://keydris.com/#faq): the canonical answers: what Keydris is and is not, where verification happens, payloads, fallback behavior
- [Security](https://keydris.com/security): how to report a security vulnerability to Keydris
- [Blog](https://keydris.com/blog): the publication front door for the thesis, research and the ongoing stream
- [Thesis](https://keydris.com/thesis): "Software is beginning to act." by Ahmed Isse, founder. The foundational argument: identity establishes who an agent is, authority establishes what it may do, and those are different questions. Covers why access is not authority, why a prompt instruction is not an authorization boundary, what ALLOW / REJECT / APPROVAL REQUIRED decide before a governed action runs, why an authorization decision of ALLOW does not mean the underlying tool succeeded, what Keydris supports today, and where the bet can fail. Cross-organization machine authority is stated in the article as direction, not current capability.
- [Research](https://keydris.com/research): original investigations and evidence
- [Start free](https://app.keydris.com/signup): create a workspace; the docs and the CLI never require it
## Full documentation index
- [Documentation index](https://keydris.com/docs): every documentation page, in dependency order