Kival

Kival is a self-hosted collaborative knowledge system built by Selemis B.V. for organizations that need durable, shared context across people and AI agents. Released as version 0.1 in September 2026, it is open-source under the Apache 2.0 / MIT dual license and written primarily in Rust.

What It Is

Kival addresses a specific organizational problem: information accumulates continuously, but the reasoning, relationships, and context that make it meaningful tend to decay or disappear. Kival gives organizations a structured place where knowledge is explicitly created, versioned, discussed, and connected — so that conclusions retain the reasoning that produced them, and earlier work remains useful as the organization grows and changes.

Unlike a general wiki or document store, Kival treats history, provenance, authorship, and relationships as first-class properties of every knowledge object. It is designed to serve both human contributors and AI agents working against the same shared knowledge base.

Core Architecture and Design Principles

Kival is built around several explicit design commitments:

• Shared knowledge: people and agents work against the same organizational knowledge rather than maintaining separate copies of context.
• Continuity: history, relationships, discussions, and provenance remain connected as knowledge develops.
• Deliberate authorship: knowledge is explicitly created and maintained with clear authorship, access, and governance.
• Durable context: knowledge remains understandable even as the people, agents, models, applications, and tools around it change.
• Self-hosted ownership: the organization retains control over its knowledge and infrastructure.

The system uses PostgreSQL as its backing store and ships a graph view for exploring connected knowledge objects alongside a detailed object view showing metadata, relationships, and local graph context.

Setup and Deployment Model

Kival is self-hosted and ships prebuilt releases for Linux x86_64 and ARM64, macOS on Apple Silicon, and Windows through WSL. Installation is handled by kivalup, a dedicated installer fetched via a single shell command. The server binary (kivald) connects to a PostgreSQL instance and exposes a web UI at port 3000 by default.

The setup path includes:

• A kivalup installer for the server binary
• Docker-based PostgreSQL for local development
• A CLI (kival) for workspace management and agent access
• A built-in demo ACME workspace for exploration before adding real organizational knowledge

Agent and Automation Integration

Kival is explicitly designed for use by AI agents alongside human contributors. Agents can use the same CLI and API key access model to inspect existing knowledge, follow relationships, and contribute new material. The README includes example agent prompts for tracing decisions, synthesizing project state into new knowledge objects, and turning operational history into follow-up knowledge.

Official SDKs are available for Rust (on crates.io) and TypeScript (on npm), both using scoped API keys and exposing the same underlying system as the CLI. The documentation covers agent access, automation, and integrations as dedicated connection categories.

Update: v0.1.3

Kival launched publicly on September 6, 2026 with version 0.1, described as a self-hosted collaborative knowledge system built for shared knowledge across people and agents. The latest release as of the available data is v0.1.3, published September 18, 2026 — indicating active early development with rapid iteration. The project is maintained by Selemis and hosted on GitHub, where community discussion takes place via GitHub Discussions. External code contributions are not currently accepted, though bug reports and feature proposals are welcomed through the issue tracker.