Connect an AI tool to your workspace.

Five steps with a workspace key. MCP first, REST for everything else. About five minutes.

When you finish you will have

  • An AI client that can remember and recall in your workspace over MCP
  • Product context added and queried over REST
  • Every request recorded as an inquiry with its evidence

Get a workspace key

Sign in at 192.168.68.60:3030. Your API key and workspace ID are shown in the dashboard under the workspace. Keys are workspace-scoped: a tool connected with one key can only see that workspace.

Set them as environment variables so the examples below work as written:

Environmentshell
export NYLOGRAM_API_KEY="nk_live_..."
export NYLOGRAM_WORKSPACE_ID="ws_..."

Connect over MCP

Nylogram exposes an MCP server. Add it to Claude Desktop, Cursor, Windsurf, or any MCP client and the tool can read and write the workspace directly, no plugin required.

MCP client configurationjson
{
  "mcpServers": {
    "nylogram": {
      "url": "https://api.nylogram.com/mcp/$NYLOGRAM_WORKSPACE_ID?key=$NYLOGRAM_API_KEY"
    }
  }
}

Once connected, the client can call remember, recall, search, forget, and stats. Use remember to add product context and recall to ask what applies before generating. The ?key= parameter is deprecated in favor of a Bearer header; see the API reference.

Add product context over REST

Send any text that describes how the product works: a requirement, a rule, a decision, a pattern description, a paragraph from a spec. Group entries with a session ID you choose, such as a workflow or a product area. Nylogram keeps the source with the entry.

Add a rulecurl
curl -X POST https://api.nylogram.com/v1/memory \
  -H "Authorization: Bearer $NYLOGRAM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "sessionId": "supplier-onboarding",
    "content": "Regional Managers cannot approve their own pricing exception requests.",
    "role": "system",
    "metadata": { "type": "rule", "source": "decision-log/2026-06-18" }
  }'
Add a rulepython
import requests

resp = requests.post(
    "https://api.nylogram.com/v1/memory",
    headers={"Authorization": f"Bearer {NYLOGRAM_API_KEY}"},
    json={
        "sessionId": "supplier-onboarding",
        "content": "Regional Managers cannot approve their own pricing exception requests.",
        "role": "system",
        "metadata": {"type": "rule", "source": "decision-log/2026-06-18"},
    },
)
print(resp.json()["id"])

Ask what applies

Query in natural language. Results come back ranked by relevance, each with the source it was drawn from, so a tool can cite what it used.

Ask what appliescurl
curl "https://api.nylogram.com/v1/memory/recall?sessionId=supplier-onboarding&q=who+can+approve+a+pricing+exception&limit=5" \
  -H "Authorization: Bearer $NYLOGRAM_API_KEY"
Ask what appliespython
resp = requests.get(
    "https://api.nylogram.com/v1/memory/recall",
    headers={"Authorization": f"Bearer {NYLOGRAM_API_KEY}"},
    params={"sessionId": "supplier-onboarding", "q": "who can approve a pricing exception", "limit": 5},
)
for item in resp.json():
    print(f"{item['score']:.2f}  {item['content']}  ({item['metadata'].get('source')})")

Inspect, export, and remove

Count what a session holds, export it as newline-delimited JSON, or remove an entry by ID. Stats and export do not count against usage.

Statscurl
curl "https://api.nylogram.com/v1/memory/stats?sessionId=supplier-onboarding" \
  -H "Authorization: Bearer $NYLOGRAM_API_KEY"
Exportcurl
curl "https://api.nylogram.com/v1/memory/export?sessionId=supplier-onboarding" \
  -H "Authorization: Bearer $NYLOGRAM_API_KEY" -o supplier-onboarding.ndjson
Removecurl
curl -X DELETE https://api.nylogram.com/v1/memory \
  -H "Authorization: Bearer $NYLOGRAM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "sessionId": "supplier-onboarding", "id": "a3f7b2e1d0c9" }'

In development

Structured Product System objects (workflows, patterns, rules, roles, components, decisions, questions) and the context tools that compose them are being built. The endpoints above are the current, supported surface; new endpoints will be added alongside them, not in place of them.

Next steps