Observation Trends

The Trends chart tracks how observation volume changes over time, broken down by observation type (architectural, performance, security, coding-pattern, etc.) and source agent. This helps you spot seasonal patterns, ramp-up curves, and potential health issues in your fleet.

Reading the Chart

The chart displays daily observations within your selected time range (7d, 30d, or 90d), with separate series for each observation type or agent.

By Observation Type

Example shape:

architectural  ▁▂▂▃▄▆▇▆▅▄▂ ← Steady ramp over 30 days
performance    ▂▁▁▂▂▁▁▁▂▁  ← Flat, infrequent
security       ▁▁▂▃▂▂▃▄▃▂  ← Variable, recent spike

Each line represents a category of observation; higher spikes mean more agents are learning about that category on that day.

By Agent

When you switch to the agent view, you see:

One line per source agent (e.g., agent-research, agent-qa)
Helps identify which agents are actively generating observations
Flat or declining lines may signal a health issue (agent disabled, scope reduced, or no matching work)

Interpretation Patterns

Pattern	Interpretation
All types trending up	Fleet learning rate is increasing; good adoption signal
One type spikes sharply	Possible campaign or reactive learning (e.g., security incident)
Type drops to zero	Category no longer being learned about; could be gap or completion
Agent line flat for >7d	Check agent health; may be disabled, out of scope, or experiencing issues
Sudden drop across all agents	Possible mass outage, access revocation, or policy change

Factors Affecting Volume

Observation counts are influenced by:

Work volume - More issues/PRs flowing through = more opportunities for agents to learn
Agent scope - Agents with broader scope will generate more observations
Memory injection tuning - If injection is disabled, agents may not generate new observations
Feedback loops - Positive feedback (confirming useful observations) encourages new learning

Time Range Selection

7 days - Short-term trends, shows daily volatility clearly
30 days - Balanced view; captures weekly cycles and gradual trends
90 days - Long-term patterns; smooths out daily noise but may hide recent spikes

Combining with Other Views

For deeper analysis, cross-reference trends with:

Coverage Heatmap - Where those observations are happening
Drift Alerts - Quality of the learning (contradictions rising alongside volume?)
Top Observations - Which learned patterns are highest-confidence

API Reference

Endpoint: GET /api/memory/analytics/trends

Query Parameters:

projectId (optional) - Filter to a specific project
since (optional) - ISO-8601 date lower bound (default: 30 days ago)

Response:

{
  byType: Array<{
    date: string  // "YYYY-MM-DD"
    observationType: string  // e.g., "architectural", "performance", "security"
    count: number
  }>
  byAgent: Array<{
    date: string
    agentId: string
    count: number
  }>
  projectId: string | null
  since: string
}

Example:

curl -X GET "https://api.rensei.ai/api/memory/analytics/trends?since=2026-05-01T00:00:00Z" \
  -H "Authorization: Bearer rsk_..."

Response:

{
  "byType": [
    {"date": "2026-05-01", "observationType": "architectural", "count": 12},
    {"date": "2026-05-01", "observationType": "performance", "count": 5},
    {"date": "2026-05-02", "observationType": "architectural", "count": 14}
  ],
  "byAgent": [
    {"date": "2026-05-01", "agentId": "agent-research", "count": 8},
    {"date": "2026-05-01", "agentId": "agent-dev", "count": 9}
  ],
  "projectId": null,
  "since": "2026-05-01T00:00:00Z"
}

Observation Types

Common observation types include:

architectural - System design and dependency patterns
performance - Optimization opportunities, bottlenecks
security - Vulnerability patterns, compliance concerns
coding-pattern - Code style, idioms, best practices
unknown - Observations without an explicit type tag

Rate Limits

The trends API enforces a 100 req/min quota per organization.

On this page