Kvendra CLI

Kvendra CLI: publication card

What it is and what problem it solves

What it is: the knowledge behind kvendra-cli (https://github.com/KvendraAI/kvendra-cli): the decisions that shaped it, its architecture and its manual, as the author recorded them while building it. The code lives in GitHub; the knowledge lives here.

This publication is a snapshot of the author's own project KB, so some entities are in Spanish. Your own KB is written in whichever language you work in.

• CLI — kvendra-cli (the kvendra binary, written in Rust). It is an MCP capability broker with a local credential vault: the broker uses your tokens on the agent's behalf instead of handing them over. An AI agent often needs to push to git, publish a package or sync a bucket; the broker runs the operation for it and constrains every call with signed allowlists, a secret-detection layer, explicit consent for destructive operations and an audit log you can check for tampering. It offers eight canonical primitives (git, github, npm, pypi, aws, http, shell and an audited escape hatch marked unsafe) and describes itself through kvendra capabilities. It is for developers and teams who work with AI agents and want an auditable boundary between the agent and their credentials.
• Do you need it? Optional on Pro: without it, the skills that only use the knowledge base work as usual. Required on Team. When installed, the skills run git, cloud and package operations through it, with your credentials kept in a local vault.
• Sister project — kvendra-platform (TypeScript; not included in this publication). It is the KB engine that anyone can run on their own machine with docker-compose up: PostgreSQL with pgvector, an MCP server, and the model of entities, transactions and change history that the Kvendra skills use. One deployment holds one team's data. Use your own embedding model, for example a local one with Ollama. It has its own publication in the catalog.
• Sister project — kvendra-skills (not included in this publication). The skills are a Claude Code plugin; the ones that run git, cloud or package operations do it through this broker. It has its own publication in the catalog.

How to use it and fork it

Read it. Anyone can read this publication in the browser.

Fork it, to work on the code with the author's knowledge. You need a Kvendra account on Pro or higher.

1. Clone the repository: git clone https://github.com/KvendraAI/kvendra-cli.git.
2. Fork this publication into your Kvendra project with the publication_fork tool in Claude Code. It becomes a new project, or is added to one you already have.
3. Review and activate. Forked entities wait in quarantine, and your AI does not use them until you review and activate them in the dashboard. Standards and patterns are activated one by one.

What your AI gains. Once you activate them, it works on the code with the author's decisions, architecture and manual in your project's knowledge base, instead of reading the code cold.

Good to know. A fork is a snapshot of one version; a new version is a new fork, and nothing is merged automatically. Private values, files and history are not copied. Self-hosted engines cannot fork.

Navigating the publication. The publication has three parts, and each one is reachable from the top navigation:

• The cover page — this card. Read it first: it tells you what is in the snapshot and how it is licensed.
• Manuals — the English book of the CLI, organised in chapters. It is the guided path through the design, the installation and the day-to-day use.
• The entity index — every published KB entity, grouped by type. Each entity page shows its content, its tags and its relations. Start with the component entity CMP-KVD-CLI, which describes the scope, stack and status of the component, and follow the relations from there.

Relations to entities that are not part of this publication are shown as "[private entity]", without an id or a title. This is intentional.

Using the code. The repository is public on GitHub:

• KvendraAI/kvendra-cli (https://github.com/KvendraAI/kvendra-cli) — install it from crates.io with cargo install --locked kvendra, or download a binary from the GitHub Releases of the repository.

The sister projects are public too: KvendraAI/kvendra-platform (https://github.com/KvendraAI/kvendra-platform) — run it locally with docker-compose up from a clone of the repository, or use the signed container image — and KvendraAI/kvendra-skills (https://github.com/KvendraAI/kvendra-skills), the plugin marketplace.

KB structure

The KB is made of typed entities. Each entity has a stable identifier of the form TYPE-PROJECT-COMPONENT-SUFFIX, Markdown content, tags, and typed relations to other entities. The types in this publication are:

| Type | What it is |
|------|------------|
| CMP | component sheet: scope, stack, status and repository of the component |
| IF | interfaces and contracts, such as the manifest of the broker primitives |
| REQ | requirements: what was asked for, why, and the acceptance criteria |
| DOC | documentation entities attached to the component |

The manual is not an entity: the CLI book is published as its own section, chapter by chapter.

The most common relations are part_of (an entity belongs to the project), affects (an entity concerns a component), fulfills (a component fulfils a requirement), depends_on and derives_from. A good reading order is: CLI book → CMP → IF → REQ → DOC.

Licenses

• CLI code (kvendra-cli): Apache-2.0.
• Content of this publication (the text of the KB entities, the manual and this card): Apache-2.0, the same license as the code.

The sister projects are not part of this publication: kvendra-platform is licensed under AGPL-3.0-only and kvendra-skills under MIT, as stated in their repositories.

Maintainer

Kvendra team — hello@kvendra.ai

Version

Content as of 2026-10-17. Component version at that date:

• CLI: kvendra 0.8.0, published on crates.io and as a GitHub Release on 2026-10-08. Status: stable alpha, under security hardening.

The publication version number is assigned by the server when the snapshot is published.

What is included and what is not

Included component:

• CLI (CMP-KVD-CLI) — kvendra-cli, the MCP capability broker with a local credential vault, together with its English book in Manuals.

For this component the publication includes its component sheet and its active entities of type CMP, IF, REQ and DOC.

Sister projects. kvendra-platform and kvendra-skills are mentioned on this card as context. Their component sheets, entities and books are not part of this publication.

Language of the entities. The card and the manual are in English. The entities are published in their original language, and part of this KB is written in Spanish.

Not included:

• Issues, tests, releases, roadmap items and architecture decision records (ISSUE, TEST, REL, ROAD and ADR) are left out by default.
• The sister projects kvendra-platform and kvendra-skills, the other components of the project, and its cross-cutting context.
• Drafts, proposals and archived entities.
• Restricted types (COST, CFG, ENV, RUN) and security or incident issues.
• Internal metadata, change history and authorship of the entities.