# RAG-Based Knowledge Store - Functional Requirements
## Overview
RAG system for CTX using **Symfony AI Store** component to accumulate, index, and retrieve knowledge about codebases.
---
## User Stories
### For Individual Developers
> "As a developer, I want the AI assistant to remember my project's context between sessions."
**Scenario**: Developer works on microservice architecture. In first session, AI studies the authorization service and
documents its API. In next session, AI can find that information via RAG search.
### For Teams
> "As a team lead, I want project knowledge to accumulate and be accessible to the entire team."
**Scenario**: New developer joins. Instead of weeks studying codebase, they get access to accumulated documentation that
AI generated.
### For AI Assistants
> "As an AI assistant, I want access to previously learned information about the project."
**Scenario**: When asked "how to write a test for the payment service," AI searches RAG for existing testing patterns
and generates code matching project style.
---
## Functional Requirements
### MCP Tools
#### Tool: `rag-store`
Store documentation and knowledge.
**Parameters:**
- `content` (required): Content to store
- `type`: Documentation type (architecture, api, testing, convention, general, tutorial, reference)
- `sourcePath`: Related source file path
- `tags`: Comma-separated tags
#### Tool: `rag-search`
Semantic search across knowledge base.
**Parameters:**
- `query` (required): Natural language search query
- `type`: Filter by documentation type
- `pathPrefix`: Filter by path prefix
- `limit`: Max results (default: 10)
- `minScore`: Minimum relevance score (default: 0.3)
#### Tool: `rag-manage`
Manage RAG entries.
**Parameters:**
- `action`: stats (more actions in future)
---
### CLI Commands
#### `ctx rag:index`
Bulk index documentation:
```bash
ctx rag:index ./docs --pattern="*.md" --recursive --type=api-docs
```
**Options:**
- `--pattern, -p`: File glob pattern (default: `*.md`)
- `--type, -t`: Documentation type tag
- `--recursive, -r`: Process subdirectories
- `--dry-run`: Preview without indexing
#### `ctx rag:clear`
Clear entries:
```bash
ctx rag:clear --force
```
#### `ctx rag:reindex`
Atomic reindex:
```bash
ctx rag:reindex ./docs --pattern="*.md" --recursive
```
#### `ctx rag:status`
View statistics:
```bash
ctx rag:status --json
```
---
## Configuration
```yaml
# context.yaml
rag:
enabled: true
store:
driver: qdrant
qdrant:
host: ${RAG_QDRANT_HOST:-localhost}
port: ${RAG_QDRANT_PORT:-6333}
collection: ${RAG_COLLECTION:-ctx_knowledge}
vectorizer:
platform: openai
model: text-embedding-3-small
transformer:
chunk_size: 1000
overlap: 200
```
---
## Integration Scenarios
### Scenario 1: Initial Documentation Index
```bash
# Index existing markdown docs
ctx rag:index ./docs --recursive --type=api-docs
# AI can now answer questions
```
### Scenario 2: Incremental Learning
```
User: "Study the payment module and document it"
AI:
1. Uses directory-list, file-read to study module
2. Forms understanding of architecture
3. Uses rag-store to save documentation
4. Returns summary to user
```
### Scenario 3: CI/CD Integration
```yaml
# .github/workflows/docs.yml
name: Update RAG Index
on:
push:
paths: [ 'docs/**' ]
jobs:
reindex:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Reindex documentation
run: ctx rag:reindex ./docs --type=api-docs
```
---
## Expected Benefits
### For AI Response Quality
| Without RAG | With RAG |
|---------------------------------------|----------------------------------------|
| "Write a test" → generic test | "Write a test" → matches project style |
| "How does auth work?" → re-reads code | "How does auth work?" → instant answer |
| Each session starts from zero | Accumulated context |
### For Development Teams
- **Onboarding** — new developers ramp up faster
- **Documentation** — created organically during work
- **Consistency** — single source of truth
- **Knowledge Preservation** — survives team changes
