← Back to Guide Docs

⚠️ IMPORTANT: All features are experimental, under active development. Use at your own risk. Customization to your workflow required. © 2026 GLG, a.s. | ← Back to Index

13. Real-World Use-Case Scenarios

Scenario A: Solo Agent with Memory (Community)

Setup: 1 AI assistant, personal use, no team.

pip install uaml-memory
uaml --accept-eula && uaml init

from uaml.facade import UAML
uaml = UAML()

# Daily workflow:
# 1. User asks about a past decision
quick = uaml.search("database migration decision", limit=3)
if quick:
    # Found it — include in response
    context = "\n".join([r.content for r in quick])
else:
    # Not stored yet — answer and store for next time
    uaml.learn(
        "Decided to use PostgreSQL over MySQL for the new project. "
        "Reasons: better JSON support, PostGIS for geo queries.",
        topic="decisions", confidence=0.95
    )

Cost: Free. ~1,000 memories. No API, no dashboard.

Scenario B: Developer Agent with Focus Engine (Starter)

Setup: 1 AI coding assistant, needs smart recall and web dashboard.

uaml = UAML()
uaml.apply_preset("standard")

# During code review — recall relevant architecture decisions:
context = uaml.recall(
    "authentication middleware patterns",
    budget_tokens=800
)
# Focus Engine returns only relevant entries within 800 tokens
# Instead of dumping 50K tokens of all code memories

# Store code review findings:
uaml.learn(
    "Code review: auth middleware lacks rate limiting. "
    "Added TODO to implement token bucket on /api/login.",
    topic="code-review", confidence=0.9
)

# Dashboard at http://localhost:8780 — visual filter tuning

Cost: €8/mo. 10K memories. REST API + dashboard + Focus Engine.

Scenario C: Power User with MCP Integration (Professional)

Setup: 1 AI agent connected via MCP, full data management.

# MCP bridge handles everything automatically:
# Agent calls memory_focus_recall → gets budget-constrained context
# Agent calls memory_store → learns from conversations

# Export for backup:
uaml.export("weekly-backup.jsonl", format="jsonl")

# Named config presets for different tasks:
# "coding" preset — strict, code-focused
# "research" preset — broad, exploratory
# Switch based on current task:
if task_type == "coding":
    requests.post("http://localhost:8780/api/v1/saved-configs/load",
        json={"name": "coding", "filter_type": "output"})
else:
    requests.post("http://localhost:8780/api/v1/saved-configs/load",
        json={"name": "research", "filter_type": "output"})

# Rules changelog tracks every config change:
GET /api/v1/rules-log  # who changed what, when, why

Cost: €29/mo. 100K memories. MCP + export/import + saved configs + audit.

Scenario D: AI Team with Coordination (Team)

Setup: 5 agents — coordinator, 2 coders, researcher, marketing. Shared knowledge.

# === MORNING: Coordinator plans the day ===
coord.claim(agent="leader", scope="daily-plan", reason="Morning planning")

# Recall yesterday's status:
yesterday = uaml.recall("what was completed yesterday?", budget_tokens=1000)

# Assign tasks:
for task in today_tasks:
    requests.post(f"{COORD}/events", json={
        "event_type": "ASSIGN",
        "agent_id": "leader",
        "target_agent": task["assignee"],
        "scope": task["scope"],
        "reason": task["description"]
    })

coord.release(agent="leader", scope="daily-plan")

# === DAYTIME: Agents work independently ===
# Each agent has Focus Engine configured for their role:
# - Coders see code + architecture (deny: marketing, sales)
# - Researcher sees research + analysis (deny: code details)
# - Marketing sees content + SEO (deny: infrastructure)

# Coder works on feature:
coord.claim(agent="coder-1", scope="src/api/v2/*", reason="New recall endpoint")
# ... codes, tests, commits ...
uaml.learn("Implemented /v2/recall with pagination support. 15 tests pass.",
           topic="code", confidence=0.95)
coord.release(agent="coder-1", scope="src/api/v2/*")

# === CREATIVE TASK: Website copy ===
# Leader broadcasts proposal request:
for agent in ["coder-1", "researcher", "marketing"]:
    requests.post(f"{COORD}/events", json={
        "event_type": "ASSIGN",
        "target_agent": agent,
        "scope": "proposal/landing-page",
        "reason": "Propose new landing page approach from your perspective"
    })

# Each agent submits proposal to shared memory:
uaml.learn("Proposal: Technical landing page with live API demo...",
           topic="proposal/landing-page", source_type="proposal")

# Leader collects all proposals:
proposals = uaml.search("proposal/landing-page", topic="proposal/landing-page")
# Summarizes for owner → owner picks best approach

# === EMERGENCY: Owner says "STOP" ===
# Bridge detects "STOP" in chat → HALT sent to all agents
# All agents pause within 30 seconds
# Owner discusses new direction → "GO" resumes work

# === END OF DAY: Sync ===
sync.export_delta(since="2026-03-16T00:00:00")
# Delta synced to all agents via Federation

Cost: €190/mo. 500K memories. Federation + RBAC + coordination + GDPR.

Scenario E: Regulated Environment (Enterprise)

Setup: Financial institution, 20+ agents, strict compliance.

# Post-quantum encryption for all stored data:
# ML-KEM-768 (NIST FIPS 203) — resistant to quantum attacks
config = {
    "encryption": {
        "enabled": true,
        "algorithm": "ML-KEM-768",
        "key_rotation": "quarterly"
    }
}

# GDPR compliance — right to be forgotten:
from uaml.compliance.auditor import ComplianceAuditor
auditor = ComplianceAuditor(store)

# Customer requests data deletion:
auditor.gdpr_erasure(customer_id="C-12345")
# → All memories referencing customer removed
# → Audit trail preserved (legally required)
# → Compliance report generated

# Full audit trail:
report = auditor.full_audit()
# Score: 94.2% | Critical findings: 0 | Recommendations: 3

# On-premise deployment — no cloud dependency:
# All data on-site, no external API calls
# Custom SLA: 99.9% uptime, 4-hour response time

Cost: Custom. Unlimited. PQC + compliance + SLA + on-prem.

← Previous Next →