Cognify

Cognify — Fleet Shared Brain

Cognify is the fleet's shared knowledge graph (vector + entity store). It runs on Seyed's Mac Mini at 100.101.29.56:8765 (Tailscale-only, never exposed publicly). All Hermes agents in Seyed's team (Appie-1, Appie-2, Spark Atlas) use it for cross-session and cross-agent memory.

Client bots get their OWN standalone Cognify instance — the fleet Cognify is NOT for client bots. Deploy Cognify on each client's Orgo machine via client-orgo-provisioning skill step 11. Client instances run locally on 127.0.0.1:8799 with the local backend (ChromaDB + networkx).

Endpoints

Recall — Query Knowledge

POST http://100.101.29.56:8765/cognify/recall
Content-Type: application/json

{
  "query": "your search query",
  "tenant": "fleet",
  "k": 6           # number of chunks to return, default 3
}

Returns: { "chunks": [...], "entities": {...}, "relations": [...] }

Always ground answers in returned chunks + entities + relations. Do not guess.

Ingest — Add New Content

POST http://100.101.29.56:8765/cognify/ingest
Content-Type: application/json

{
  "source": "appie-1",               # REQUIRED — who is ingesting (field is "source" NOT "agent")
  "namespace": "workspace:appie-1",   # REQUIRED — namespace for organization
  "text": "...",                      # REQUIRED — the content (field is "text" NOT "content")
  "title": "optional title"           # OPTIONAL — overrides auto-title from first line
}

Response: { "doc_id": "...", "title": "...", "tenant": "fleet", "namespace": "...", "chunks": N, "entities": N, "relations": N, "extracted": true }

Critical details (verified by testing):

• Field name is "text" — NOT "content" (older docs may be wrong)
• Field name is "source" — NOT "agent"
• Do NOT use "path" — that's for server-side file reading; our agents don't share a filesystem with the server
• Documents ~10KB need up to 120s timeout (entity extraction is compute-heavy)
• Returns chunk count, entity count, and relation count

Python Patterns

Ingest (preferred — handles long content)

import json, urllib.request

with open(file_path) as f:
    text = f.read()

payload = {
    "source": "appie-1",
    "namespace": "workspace:appie-1",
    "text": text
}

data = json.dumps(payload).encode()
req = urllib.request.Request(
    "http://100.101.29.56:8765/cognify/ingest",
    data=data,
    headers={"Content-Type": "application/json"},
    method="POST"
)
with urllib.request.urlopen(req, timeout=120) as resp:
    result = json.loads(resp.read())
    print(f"Ingested: {result['chunks']} chunks, {result['entities']} entities, {result['relations']} relations")

Recall

import json, urllib.request

payload = {"query": "your search", "tenant": "fleet", "k": 6}
data = json.dumps(payload).encode()
req = urllib.request.Request(
    "http://100.101.29.56:8765/cognify/recall",
    data=data,
    headers={"Content-Type": "application/json"},
    method="POST"
)
with urllib.request.urlopen(req, timeout=30) as resp:
    results = json.loads(resp.read())
    # results["chunks"] — text chunks ranked by relevance
    # results["entities"] — named entities extracted
    # results["relations"] — relationships between entities

Health Check

GET http://100.101.29.56:8765/health

Returns {"status": "ok", "index_ready": true, "total_chunks": N, ...}

Key Differences From Earlier Documentation

Notes

• Tailscale-only endpoint. Do not expose publicly.
• Fleet endpoint path: /cognify/recall, /cognify/ingest (nginx reverse proxy)
• Client-local endpoint path: /recall, /ingest (NO /cognify/ prefix)
• Namespace convention: workspace:<agent-name> (e.g. workspace:appie-1, workspace:zeus).
• Tenant: "fleet" for fleet Cognify, bot-specific (e.g. "zeus") for client instances.
• Entity extraction runs automatically on ingest — no separate call needed.