Use Cases

# 1. Prepare training data from your codebase eldric> /train prepare ./src --output training-data.jsonl # 2. Fine-tune a model on your code eldric> /train create --base llama3.1:8b --data training-data.jsonl --epochs 3 # 3. Use your custom model eldric> /model my-codebase-expert eldric> How does authentication work in this project? AI: Based on your codebase, you use JWT tokens with refresh rotation. The auth middleware is in src/middleware/auth.ts, and tokens are validated using the verifyToken() function from src/utils/jwt.ts...

# Index your documentation eldric> /learn ./company-docs --recursive [Learn] Indexing 847 documents (312 MB)... [Embed] Creating vectors with nomic-embed-text... [Done] Knowledge base ready! 12,456 chunks indexed. # Ask questions eldric> What's our incident response process for P1 issues? AI: Based on your runbooks (incident-response.md): 1. Acknowledge in #incidents Slack channel within 5 minutes 2. Page on-call engineer via PagerDuty 3. Create incident ticket with template from JIRA 4. Begin investigation, update every 15 minutes 5. Post-mortem required within 48 hours [Source: docs/runbooks/incident-response.md, lines 23-45]

# Create a merged model eldric> /merge create Select models: 1. codellama:13b (code generation) 2. mistral:7b (reasoning) 3. my-domain-lora (your expertise) Strategy: SLERP Weights: 0.4 / 0.3 / 0.3 [Merge] Processing tensors... [Done] Created: my-code-reasoning-expert (14.2 GB) # Your merged model combines all three capabilities

# Deploy edge with TLS ./eldric-edge --port 443 \ --cert /etc/ssl/company.pem --key /etc/ssl/company.key \ --routers http://router:8881 # Register team clients with different limits curl -X POST https://ai.company.com/api/v1/clients/register \ -d '{"name": "Marketing Team", "rate_limit_rpm": 500}' # Returns: api_key: eld-marketing-xxx curl -X POST https://ai.company.com/api/v1/clients/register \ -d '{"name": "Sales Team", "rate_limit_rpm": 1000}' # Returns: api_key: eld-sales-xxx # Configure OpenWebUI OPENAI_API_BASE_URL=https://ai.company.com/v1 OPENAI_API_KEY=eld-marketing-xxx

Customer: "How do I reset my password?" [Edge] Received request, rate limit OK [Router] AI classification: SIMPLE_FAQ [Router] Selected: worker-fast-01 (llama3.2:3b) [Worker] RAG search: "password reset" → 3 relevant docs [Worker] Generated response with citations Response: "To reset your password: 1. Go to login page and click 'Forgot Password' 2. Enter your email address 3. Check your inbox for reset link (expires in 1 hour) 4. Click link and create new password [Source: help-center/account/password-reset.md]" Latency: 145ms | Tokens: 89 | Cost: $0.00003

# Submit research goal to swarm orchestrator curl -X POST http://swarm:8885/api/v1/goals -d '{ "description": "Systematic review: ML in drug discovery 2020-2024", "mode": "supervised", "sources": ["pubmed", "arxiv", "nature", "science"] }' # Swarm coordinates 8 agents in parallel: # Searcher → Explorer → Coder → Database → Planner # Result: 523 papers analyzed, 12 themes identified in 2 hours

# Define experiment goal with constraints eldric-swarm> /goal "Design CRISPR protocol for gene ABC123" --mode advisory --budget 5000 --timeline 14d # Swarm generates: # ├── Literature review (12 relevant protocols) # ├── Guide RNA design (3 candidates + off-target analysis) # ├── Optimized protocol for available equipment # ├── Materials list with vendor quotes # ├── Python analysis scripts for flow cytometry # └── Safety documentation (IBC compliance) # Each step requires researcher approval for audit trail

# Morning: Check for new relevant papers eldric-swarm> /goal "Check arxiv and pubmed for my thesis topics" → Found 7 new papers, added to queue with summaries # Afternoon: Writing assistance eldric-swarm> /goal "Write methods for chapter 3 using lab notebooks" → Generated 2,400 words with 23 citations # Evening: Committee meeting prep eldric-swarm> /goal "Statistical analysis on dataset DS-2024-Q3" → 8 publication-ready figures + ANOVA results

# Start grant development swarm eldric-swarm> /goal "NIH R01 for computational drug discovery" --deadline 2024-06-01 --agency NIH Swarm agents (6 coordinated): → Searcher: 3 matching FOAs found (94% match) → Explorer: 47 related funded R01s analyzed → Database: Preliminary data from lab publications → Coder: SF424 formatting applied → Planner: Timeline and milestones created → Reviewer: Simulated study section critique Progress: Specific Aims ready, Background 60% complete Budget auto-calculated from HR + equipment + F&A rates

# Upload meeting recording curl -X POST http://media:8894/api/v1/video/transcribe \ -F "file=@standup-2024-01-30.mp4" \ -F "diarize=true" \ -F "summarize=true" # Result { "transcript": [...], "speakers": ["Alice", "Bob", "Charlie"], "summary": "Sprint review discussed feature X completion...", "action_items": [ {"assignee": "Bob", "task": "Deploy to staging by Friday"}, {"assignee": "Alice", "task": "Review PR #423"} ], "duration_seconds": 1847, "indexed": true }

# Configure Signal account curl -X POST http://comm:8895/api/v1/comm/accounts \ -H "Content-Type: application/json" \ -d '{ "protocol": "signal", "name": "Secure Ops Channel", "credentials": { "phone_number": "+1234567890", "signal_cli_path": "/usr/local/bin/signal-cli" }, "ai_mode": "disabled" # No AI for sensitive comms }' # All messages stored locally with E2E encryption # Searchable via RAG but never sent to cloud