Practical overview
When it fits
Use Akb when an MCP-connected agent needs a persistent organizational knowledge store with structured content, files, hybrid retrieval and explicit relationships, and your team can operate the documented self-hosted stack.
Who it is for
- Teams that want MCP-connected agents to read and update shared organizational knowledge.
- Operators who can self-host a PostgreSQL-based service and configure its agent access.
- Developers using named MCP clients such as Claude Code, Claude Desktop, Cursor, Windsurf, Cline or Continue.
Documented uses
- Store and retrieve organizational documents, structured tables and files for agents.
- Search knowledge using semantic and keyword retrieval.
- Connect knowledge objects through URI-based relationships.
- Expose a shared read/write knowledge vault to MCP-aware agents.
Requirements & limits
The README describes a public demo that can be browsed and searched without signup. It is reset weekly, is described as having no uptime, privacy or data guarantees, and should not be used for real or sensitive information. The default agent connection uses a Personal Access Token; an optional Keycloak-backed OAuth path is also documented.
- The native self-hosted setup requires Docker Compose 2.24.4 or later and a new Compose project with a never-used database for new installations.
- Dense search requires an OpenAI-compatible embedding endpoint; without embeddings, pgvector and Qdrant fall back to BM25-only lexical search.
- AKB’s core is a read/write store rather than a built-in summarizer or knowledge consolidator; event-driven follow-on workflows require external consumers, with Redis configured for Redis Stream publishing.
- The README reports 98.4% Recall@5 on LongMemEval-S without a reranker and notes that embedding models differ across systems.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Prepare a native self-hosted installation
Build AKB images, prepare installation configuration and start the documented Docker Compose stack.
Before you start
- Docker Compose 2.24.4 or later
- A new Compose project and a never-used database for a new installation
- An OpenAI-compatible embedding endpoint for dense search
- Provider and public URL settings, credentials and independent secrets
Steps
Build the backend image from the checkout and build the frontend and PostgreSQL images with Docker Compose.
Copy the example application and secret configuration files into the native configuration directory, then edit them with provider, public URL and credential settings.
Generate a local session keyset and prepare the native configuration using the documented CLI commands, installation tenant ID and namespace.
Start the stack with the standard and native Compose files using the same project and configuration files.
Expected outcome
The documented native AKB stack is started; the next documented step is provisioning a local recovery administrator.
Guide limitations
- The quick-start instructions require a new Compose project and never-used database; existing installations must follow the separate pre-upgrade preservation procedure.
- The guide does not cover recovery-administrator provisioning.
- Without an embedding endpoint, pgvector and Qdrant degrade to BM25-only lexical search.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- dnotitia/akb · Official homepage title · retained in its source language