# Mithril Graph & Ontology database

> The Knowledge Graph & Ontology database of Mithril (formerly the
> kotobase.net graph plane; the workbench moved to graph.kotoba.cloud on
> 2026-09-16). Tenant-isolated graphs of datoms over immutable IPLD
> snapshots, every entity tagged with an ontology type, queried in Datalog
> (kotobase dialect, Datomic-shaped EDN), Cypher, SPARQL, Gremlin bytecode,
> or bounded read-only SQL. Content identity is a CID.

Workbench (sign-in): https://graph.kotoba.cloud/
API origin: https://api.mithril.fund/v1/database (session cookie or Biscuit; anonymous is 401)
Read-only query lane for tool-calling models: https://api.mithril.fund/v1/graph/query · schema https://api.mithril.fund/v1/graph/tools
Public operator: Kotoba Labs Inc. Contact: support@kotoba.cloud.

## Start in one command (MCP)

MCP server: https://graph.kotoba.cloud/mcp (streamable HTTP, stateless; descriptor
https://graph.kotoba.cloud/.well-known/mcp.json). With the two values from the
workbench in the environment:

```sh
claude mcp add --transport http kotoba-graph https://graph.kotoba.cloud/mcp --header "Authorization: Bearer $KOTOBA_DATABASE_TOKEN" --header "X-Kotobase-Tenant: $KOTOBA_TENANT_ID"
```

Tools: graph_list_databases · graph_create_database · graph_db_stats ·
graph_transact · graph_query (datalog | cypher | sparql | sql) · graph_usage ·
graph_docs (no credential). A tool result is the HTTP route's own answer.

## Agent entry points

- Executable acceptance path (workspace → agent token → Biscuit → create database → transact → query): https://graph.kotoba.cloud/agent-quickstart.md
- Expanded model context (data model, every method, limits, evidence): https://graph.kotoba.cloud/llms-full.txt
- Agent card: https://graph.kotoba.cloud/.well-known/agent.json
- Human docs: https://docs.mithril.fund/graph/ · error codes: https://docs.mithril.fund/reference/errors/
- Health (no data access): https://graph.kotoba.cloud/health

## What needs a person

Sign-in (passkey / SIWE / own key), creating a workspace (`t_…`), and minting
the agent token (a tenant service account, `kb_sa_…`, shown once) happen in
the workbench. An agent cannot sign in and must never be handed a passkey,
wallet, or card. With `KOTOBA_TENANT_ID` and `KOTOBA_DATABASE_TOKEN` in its
secret store the agent sends `Authorization: Bearer $KOTOBA_DATABASE_TOKEN`
on every call (or attenuates it to a 15-minute, one-database Biscuit at
`POST https://auth.mithril.fund/v1/biscuit/token`) and needs nobody else.

## Minimal verification

```sh
curl -fsS https://api.mithril.fund/v1/graph/tools     # public tool schema
# then agent-quickstart.md steps 3–4 (or the MCP tools): createDatabase → transact → q
```

Accept only a `datomic.q` response whose rows contain the value you wrote.
A health 200, a tool schema, or a minted Biscuit is not evidence of a write.

## Data model in one paragraph

A database is a set of datoms `[e a v tx added]` behind a signed immutable
commit DAG (CID). Entities are maps of attributes; an attribute whose value is
another entity is an edge. `:ontology/type` is the type tag the ontology
inspector reads (Cypher node label, Gremlin `hasLabel`). Transactions are EDN
vectors (`tx_edn`); a write publishes only after a conditional head
compare-and-swap wins (HTTP 409 `retryable: true` otherwise).

## Claim boundaries

Immutable snapshots and CID identity make stored content portable and
independently checkable; they do not make hosted write availability
provider-independent. Limits, measured latency, and the exact allowlists are
in llms-full.txt. Do not infer customer traction or an SLA from this document.

## The rest of the stack

- Kotoba, the language: https://kotoba-lang.org/llms.txt
- Mithril identity, inference and packages: https://mithril.fund/llms.txt