build-embeddings-cache

name	build-embeddings-cache
description	Generates the MITRE ATT&CK technique embedding cache (technique_embeddings.json). Required after updating enterprise-attack.json or if the cache is missing or corrupt. Takes ~3 minutes and makes ~274 API calls to OpenRouter. Run after /update-data completes a MITRE refresh.
allowed-tools	Bash(python3:*) Bash(ls:*) Read
compatibility	Requires OPENROUTER_API_KEY in .env and internet access. Verify the embedding model name is still a valid free-tier model on openrouter.ai before running.

Build Embeddings Cache

Generates chatbot/data/technique_embeddings.json (~45 MB) from chatbot/data/enterprise-attack.json.

Pre-flight

# Confirm MITRE data exists
ls -lh "$(git rev-parse --show-toplevel)/chatbot/data/enterprise-attack.json

# Check current embedding model is still available on OpenRouter
# Visit https://openrouter.ai/models and search for the model name below
# Free-tier models rotate — verify before running if it has been >30 days
grep -r "embed" "$(git rev-parse --show-toplevel)/chatbot/modules/mitre_embeddings.py | grep "model"

Run

cd "$(git rev-parse --show-toplevel)"
python3 -c "
from chatbot.modules.mitre_embeddings import build_technique_embeddings, save_embeddings_json
from chatbot.modules.mitre import MitreHelper
from chatbot.modules import embeddings
mitre = MitreHelper(use_local=True)
cache = build_technique_embeddings(mitre, embeddings)
save_embeddings_json(cache, 'chatbot/data/technique_embeddings.json')
"

Expected outcome

• Runtime: ~3 minutes (~274 requests, rate-limited)
• Output: chatbot/data/technique_embeddings.json — should be ≥40 MB
• Technique count: ≥835 embeddings

Failure handling

Symptom	Fix
OPENROUTER_API_KEY missing	Add to .env
Model 404 / not found	Update model name in chatbot/modules/mitre_embeddings.py
Rate limit errors	Wait and re-run — the script resumes from where it left off
Output file <10 MB	JSON truncated; delete and re-run