Ops Deck Lite

Build two small local services before you build a giant agent dashboard.

Goal

Give the agent two capabilities that pay off immediately:

semantic code search
a reusable prompt library

Component 1: Semantic code search

Search code by meaning, not only exact text.

Useful for queries like:

authentication middleware
retry logic
database pooling
error handling

Core design

chunk source files
generate embeddings locally or through a provider you control
store chunks, metadata, and optional summaries in a small database
expose one search endpoint with a few modes: hybrid, code, summary

Component 2: Prompt library

Store prompts as reusable templates instead of rewriting them every session.

Each prompt should have:

title
category
body
optional tags
optional variables

Why this stack works

cheap to run
private by default
easy to back up
useful to both humans and agents
much smaller blast radius than a full operations suite

Minimal setup guidance

Use any stack you like, but keep the shape simple:

one code-search API
one prompt-library API
one small database per service or shared storage if you prefer
one lightweight process manager or service supervisor

Agent integration pattern

Teach the agent two habits:

Check semantic code search before grepping dozens of files.
Check the prompt library before drafting a fresh prompt from scratch.

Suggested API shape

Code search

POST /api/search
GET /api/health
POST /api/index

Prompt library

GET /api/prompts
GET /api/prompts/:id
POST /api/prompts
PUT /api/prompts/:id
DELETE /api/prompts/:id

Verification

A good smoke test:

query a concept not literally present in source code
confirm semantic search still finds the right area
retrieve a known prompt by category and reuse it successfully

Anti-pattern

Do not start with a full control plane, analytics wall, sprint board, and seven side panels. Start with the two tools that actually save tokens and time.