Organizational Knowledge System
Run Demarkus at organization scale: a broker-fronted universe of worlds that whole teams join with one command, authenticate against with OIDC, and reach through the same MCP tools their agents already use.
This is the tier above the team knowledge base. Users sign in through your identity provider; the gateway checks each world’s access and write scope. Nobody copies a raw world token by hand.
What you’ll have
demarkus-knowledge: one process hosting worlds, OIDC login, and MCP gateways.- One or more worlds: isolated catalogs, each with its own GCS bucket, policy, and access rules.
- A hub: the world holding shared conventions and the cross-world graph, derived by the knowledge server.
- A library: the reading room for humans, behind the same login, over the same worlds.
- No hand-distributed tokens: users run
/knowledge-joinand authenticate in the browser. - Full version history, capability auth, and audit logging on every world, same as any Demarkus server.
Architecture
Coding agents Library
│ │
└──── MCP + OIDC ───┘
│
demarkus-knowledge
gateways + worlds + graph
│
┌─────────┼─────────┐
▼ ▼ ▼
root planner storage
one GCS bucket per world
Gateway calls reach worlds in process. mark_worlds lists readable worlds;
mark_lookup_all searches them in one call and reports partial failures. Lookup
matches tags and titles, or section text with match=body; budget can include
focused passages and related documents.
Direct QUIC clients select a world with TLS SNI. Capability tokens are for direct clients; gateway writes use the signed-in identity’s world grant.
High availability
Durable state lives in GCS. World writes commit through a compare-and-swap head and commit log; replicas catch up from the bucket. No persistent volumes or writer leader election. The chart requires at least two replicas and keeps one available during voluntary disruption.
The server initializes an empty world on first start and seeds its write policy. The floor world’s policy applies across the system; each world can tighten it. Writes are checked for secrets and personal data. GCS access uses workload identity; direct-client tokens, when needed, mount from Secrets.
For the design rationale behind worlds, hubs, and memories, see Creating an automated knowledge universe.
Stand up the system
Use the knowledge-system Helm chart: one values file declares the worlds, OIDC gateway, and optional library. The deployment repository provides the GKE and GitOps setup.
The deployment publishes a system policy and project template to the hub’s well-known path (mark://root/.well-known/demarkus/), so joining agents discover the conventions of your system, required tags and project layout included, without you documenting them separately.
Join as a user
Once the broker is up, anyone on the team joins from Claude Code, Cursor, OpenCode, or pi. No local server or capability token is needed: the plugin connects the broker through the agent’s MCP OAuth flow.
1. Install the plugin
Claude Code:
/plugin marketplace add latebit-io/demarkus
/plugin install demarkus-knowledge@demarkus
OpenCode:
curl -fsSL https://raw.githubusercontent.com/latebit-io/demarkus/main/plugins/opencode-knowledge/install.sh | bash
Restart OpenCode before continuing.
pi:
pi install npm:pi-mcp-adapter
pi install git:github.com/latebit-io/demarkus-pi-knowledge
Cursor, with the repository added as a marketplace from the plugins panel (or symlinked from a checkout into ~/.cursor/plugins/local/):
Customize → Add Marketplace → Import from GitHub → https://github.com/latebit-io/demarkus (scope: User) → Import
demarkus-knowledge → Add
Restart Cursor before continuing.
2. Join the system
/knowledge-join https://knowledge.example.com
Pass the full https:// URL of the MCP gateway. The command registers it in your
agent and starts the browser-based OAuth flow. You log in through your identity
provider; the gateway enforces access per world.
In OpenCode, restart after joining. In V2, open /mcps, select the server, and
sign in; in V1, complete OAuth when prompted or run opencode mcp auth <slug>.
In Cursor, restart and complete OAuth under Settings → MCP.
3. Navigate and work
/knowledge
The plugin steers agents to consult the system first on shared subjects. Ready
material lands through /promote and a human gate, with the system’s policy
enforced at write time.
A personal memory and a knowledge system compose in the same agent. Draft in your memory; promote what others should rely on.
Live example
knowledge.demarkus.io is a live knowledge system running the reference deployment, with a root hub and a published system policy. Point /knowledge-join at it to see the join flow end to end:
/knowledge-join https://knowledge.demarkus.io
Related
- How knowledge flows: the agent-driven capture-and-promote loop, personal memory up to the shared catalog
- Team knowledge base: the single-server tier below this
- Agent memory: personal agent memory that composes with a knowledge system
- Creating an automated knowledge universe: the worlds / hubs / memories design
- Ecosystem: plugins and tools that speak Demarkus