llms.txt and llms-full.txt

The docs site generates three machine-readable digests at build time via the starlight-llms-txt plugin. All three follow the llmstxt.org proposal.

The three URLs

URL	Format	Use for
`/llms.txt`	Manifest of pages + short descriptions	Agent picks which pages to fetch on demand
`/llms-full.txt`	All EN docs concatenated as markdown	One-shot ingestion of the whole corpus
`/llms-small.txt`	Filtered subset (core concepts only)	Small-context models (≤ 32k)

When to use which

Pull llms-full.txt into the model’s prompt or system prompt at the start of the session. Works for Opus, GPT-5, Gemini 2.5 — any model with ≥ 128k context.

curl -s https://docs.namidb.com/llms-full.txt > namidb-context.md
# Then feed namidb-context.md into your prompt / system message

Pull llms.txt to get the page manifest, then have the agent fetch individual .md URLs on demand (every page also serves raw markdown at the same URL with .md appended).

curl -s https://docs.namidb.com/llms.txt
# Inspect the page list, then for any page e.g.:
curl -s https://docs.namidb.com/en/cypher/write-queries.md

Use llms-small.txt when the model has ≤ 32k context. It keeps the conceptual core (the bucket, snapshots, namespaces, Cypher subset, SDKs) and drops the deep RFCs and operational reference.

curl -s https://docs.namidb.com/llms-small.txt | wc -w
# roughly ~12k words

What’s in `llms-full.txt`

The English locale, organized by section:

Get started (5 pages)
Concepts (6 pages)
Cypher reference (5 pages)
SDK reference (4 pages: Python, Rust, CLI, HTTP)
Operations (12 pages: config, URI grammar, 6 storage backends, self-host, observability, backup, tuning)
Cloud (3 pages)
Internals (RFCs) (17 RFCs)
Community (4 pages)
Changelog

The Spanish locale is excluded from the LLM digest by design — having both locales in one corpus just dilutes the signal for agents that don’t speak Spanish anyway. The Spanish HTML site stays fully browsable.

Provider-specific ingestion notes

Claude Code

If you’re using Claude Code in a repo that touches NamiDB, install the Claude Code skill instead — it’s a tighter, in-context summary tuned for IDE workflows. Use llms-full.txt only when the skill isn’t installed or you’re working with a different LLM front-end.

Anthropic API directly

import anthropic, urllib.request

namidb_context = urllib.request.urlopen(
    "https://docs.namidb.com/llms-full.txt"
).read().decode()

client = anthropic.Anthropic()
resp = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=2048,
    system=[
        {
            "type": "text",
            "text": namidb_context,
            "cache_control": {"type": "ephemeral"},
        }
    ],
    messages=[{"role": "user", "content": "Write a Cypher query that …"}],
)

The cache_control block enables prompt caching so subsequent calls in the same session don’t re-pay the ingestion cost.

Cursor / Cody / Continue

Most editor-embedded agents support a “rules” or “context” file. See Cursor, Codex, AGENTS.md for drop-in templates that reference llms-full.txt.

Codex (OpenAI)

from openai import OpenAI
import urllib.request

namidb_context = urllib.request.urlopen(
    "https://docs.namidb.com/llms-full.txt"
).read().decode()

client = OpenAI()
resp = client.responses.create(
    model="gpt-5",
    instructions=f"You have full NamiDB documentation in context:\n\n{namidb_context}",
    input="Write a Cypher query that …",
)

How it’s built

import starlightLlmsTxt from 'starlight-llms-txt';

starlight({
  plugins: [
    starlightLlmsTxt({
      projectName: 'NamiDB',
      description: 'Cloud-native graph database…',
      exclude: ['es/**'],
      customSets: [
        { label: 'Cypher reference', paths: ['en/cypher/**'] },
        // …
      ],
    }),
  ],
})

Re-runs on every pnpm build and on every Vercel push. Always matches the live site.