ACE Observer

ML-Powered

The senior developer that never sleeps - intelligent AI monitoring that learns and improves over time

Intelligent Learning System

Observer actively learns from every interaction

AI Output→Observer→Observations→Memories

High-value observations automatically become searchable memories

1. Monitor

AI output analyzed in real-time

2. Detect

Patterns, conflicts, errors found

3. Learn

Auto-promote to memories

4. Improve

Learning Agent finds patterns

Key Features

Auto-Learn

High-confidence observations automatically become searchable memories for future recall

Learning Agent

Background AI reviews observations periodically and extracts cross-session patterns

Centralized Storage

All observations route to a single configurable namespace - no more scattered logs

Feedback Learning

Your feedback improves observations - helpful items get promoted, noise gets filtered

Configuration

Customize Observer behavior from the dashboard

Navigate to Dashboard → Observer → Configuration to adjust settings:

Observation Namespace

Where all observations are stored (centralized)

Observer Role

developer, architect, pm, qa, security, general

Confidence Threshold

Minimum confidence (0-100%) to persist observations

Auto-Learn Threshold

Confidence needed to auto-promote to memories (default 85%)

Feature Toggles

• Inject Context - Add project memories to AI prompts
• Validate Responses - Analyze AI outputs for issues
• Persist Observations - Store observations in database
• Auto-Learn - Promote high-confidence observations to memories

Learning Agent

Background AI that extracts patterns from observations

The Learning Agent runs periodically (default: every 6 hours) to find patterns across your observations:

Repeated Issues

Same bug appearing multiple times

AI Mistakes

AI consistently making the same error

Best Practices

Successful patterns worth standardizing

What the Learning Agent Does

Reviews unprocessed observations from the last 12 hours
Uses AI to find patterns across observations
Creates memories from high-confidence patterns
Proposes decisions from best practices found
Detects decision drift (AI behavior conflicting with recorded decisions)

View Learning Agent status and trigger manual runs from Dashboard → Observer

What Observer Catches

Types of issues Observer detects automatically

error_missed

Ignored Errors

AI mentioned an error but dismissed it without resolving

pattern

Circular Reasoning

Same topic discussed multiple times without resolution

conflict

Decision Conflicts

AI output contradicts recorded architectural decisions

pattern

Flip-Flop

AI reversing a recent decision within a few turns

security

Security Concerns

Potential vulnerabilities, hardcoded credentials, unsafe code

MCP Tools

Use Observer from Claude Code, Cursor, and other AI tools

ace_observe

Send content for real-time observation:

ace_observe({
  content: "AI output to analyze",
  context: "What you're working on",
  role: "developer"
})

ace_review_observations

Review recent observations for learning opportunities:

ace_review_observations({
  filter_type: "conflict",    // Optional: conflict, error_missed, etc.
  filter_severity: "warning", // Optional: info, warning, error, critical
  limit: 10                   // Max results
})

ace_trigger_learning

Manually trigger the Learning Agent:

ace_trigger_learning()

// Returns:
// {
//   "observations_reviewed": 15,
//   "patterns_found": 2,
//   "memories_created": 1,
//   "decisions_proposed": 1
// }

Observer Roles

Choose a role to customize what Observer looks for

developer

Code quality, patterns, best practices, technical debt

architect

System design, architecture decisions, dependencies

pm

Requirements alignment, scope creep, priorities

qa

Test coverage, edge cases, quality issues

security

Vulnerabilities, access control, data protection

general

All-purpose review across all categories

Feedback Loop

Help Observer learn what matters to you

Your feedback directly improves Observer's learning:

Helpful

Boosts confidence by 10% and attempts to promote observation to memory (with lowered threshold)

Not Helpful

Prevents auto-promotion to memory - observation won't become a searchable insight

Best Practices

Set a Dedicated Namespace

Configure a specific namespace (e.g., "observer-logs") to centralize all observations

Enable Auto-Learn

Let high-value observations automatically become searchable memories

Provide Feedback

Rate observations to help Observer learn your preferences

Monitor Learning Agent

Check the dashboard to see patterns found and decisions proposed

Choose the Right Role

Use security role for auth code, architect for design decisions, etc.