Statistics API

Endpoints for querying agent activity analytics, growth metrics, daily work journals, and AI-generated daily summaries.

Daily Journal

Retrieve a consolidated view of all agent activity for a specific day.

GET /api/v1/statistics/daily-journal?date=YYYY-MM-DD&agent_id=optional

Parameters

Parameter	Type	Required	Description
`date`	string	Yes	Date in `YYYY-MM-DD` format
`agent_id`	string	No	Filter by specific agent ID

Response

{
  "code": 0,
  "data": {
    "date": "2026-05-31",
    "overview": {
      "total_sessions": 5,
      "total_tokens": 42000,
      "total_cost": 0.35,
      "tool_call_count": 28,
      "approval_count": 2,
      "cron_run_count": 1,
      "kanban_event_count": 3,
      "sessions_by_source": {
        "web": 3,
        "telegram": 1,
        "api": 1
      }
    },
    "sessions": [
      {
        "id": "abc-123",
        "title": "Code review session",
        "source": "web",
        "started_at": "2026-05-31T09:15:00Z",
        "total_tokens": 12000,
        "total_cost": 0.10
      }
    ],
    "approvals": [],
    "cron_runs": [],
    "kanban_events": [],
    "timeline": [
      {
        "type": "session",
        "time": "2026-05-31T09:15:00Z",
        "title": "Code review session",
        "detail": { "source": "web", "tokens": 12000 }
      }
    ]
  }
}

Data Sources

The journal aggregates data from 6 existing sources without any additional storage:

Source	Data
Chat	Session metadata (title, source, timestamps)
Message	Token counts and cost per session
ApprovalRecord	Human approval events
CronRunModel	Scheduled task executions
KanbanTaskEventModel	Kanban board events
EventLog	Tool call counts (file-based)

Error Responses

Code	Description
400	Invalid date format (must be `YYYY-MM-DD`)
400	Missing `date` parameter

Daily Wrap (AI Summary)

Get an AI-generated natural-language summary of a day’s activity, including keywords and next-day suggestions. Results are cached in SQLite to minimize LLM costs.

GET /api/v1/statistics/daily-wrap?date=YYYY-MM-DD

Parameters

Parameter	Type	Required	Description
`date`	string	Yes	Date in `YYYY-MM-DD` format

Response

{
  "code": 0,
  "data": {
    "date": "2026-06-27",
    "summary": "Productive day focused on code review and bug fixes. Completed 5 sessions across web and Telegram channels.",
    "keywords": ["code review", "bug fix", "telegram"],
    "suggestions": ["Continue the refactoring started in session #3", "Review pending approvals"],
    "generated_at": "2026-06-27T23:05:00Z",
    "cached": true
  }
}

Regenerate

Force a fresh AI summary, bypassing the cache:

POST /api/v1/statistics/daily-wrap/regenerate?date=YYYY-MM-DD

Returns the same response structure as the GET endpoint with cached: false.

Requirements

A Lite Model must be configured in Settings (used for low-cost summary generation)
If no activity exists for the date, returns reason: "no_activity" with summary: null
If no Lite Model is configured, returns reason: "lite_model_not_configured" with summary: null

How It Works

Aggregates data from the same 6 sources as Daily Journal (sessions, tokens, approvals, cron runs, kanban events, cost)
Builds a structured prompt and sends it to the configured Lite Model
Parses the LLM response (supports JSON and plain-text fallback)
Caches the result in a dedicated daily_wrap_cache SQLite table (one row per date)

Per-Agent Usage Analytics

Break down token consumption and cost by individual agent — see which agents consume the most resources, with 7-day trend sparklines.

GET /api/v1/statistics/usage/by-agent?days=7

Parameters

Parameter	Type	Required	Description
`days`	integer	No	Number of days for sparkline data (default: 7, max: 30)

Response

{
  "success": true,
  "data": {
    "agents": [
      {
        "agentId": "builtin-general",
        "name": "General Assistant",
        "avatar": "icon:general",
        "totalTokens": 125000,
        "totalUsd": 1.25,
        "totalCalls": 42,
        "sessions": 15,
        "percentTokens": 65,
        "percentUsd": 72,
        "sparkline": [
          { "date": "2026-06-03", "tokens": 18000, "usd": 0.18 },
          { "date": "2026-06-04", "tokens": 22000, "usd": 0.22 }
        ]
      }
    ],
    "total_agents": 3,
    "grand_total_tokens": 192000,
    "grand_total_usd": 1.74
  }
}

Key Features

Results sorted by total USD cost (highest first)
Percentage breakdown shows each agent’s share of total consumption
Sparkline data enables 7-day SVG trend visualization in the UI
Automatically resolves agent names and avatars from the Agent registry
The AgentUsageCard component auto-hides when there is only 1 agent (no comparison value)

Session Execution Trace

Retrieve the complete execution trace for a session — tool calls, LLM invocations, errors, human feedback events, and memory operations — structured for timeline replay.

GET /api/v1/statistics/session/{session_id}/trace

Response

{
  "code": 0,
  "data": {
    "session_id": "sess-abc123",
    "metadata": {
      "user_id": "user-1",
      "agent_id": "builtin-general",
      "task_type": "chat",
      "trace_id": "trace-xyz"
    },
    "outcome": "success",
    "start_time": 1720000000.0,
    "end_time": 1720000030.0,
    "duration_ms": 30000,
    "task_input": "Help me refactor this module",
    "output": "Done! I've refactored the module into 3 files.",
    "tool_calls": [
      {
        "sequence": 1,
        "tool_name": "read_file",
        "start_time": 1720000002.0,
        "end_time": 1720000003.5,
        "duration_ms": 1500,
        "success": true,
        "error": null,
        "input_data": { "path": "/src/module.py" },
        "output_summary": "Read 200 lines"
      }
    ],
    "llm_calls": [
      {
        "sequence": 1,
        "start_time": 1720000001.0,
        "end_time": 1720000005.0,
        "model_name": "claude-sonnet-4-20250514",
        "prompt_preview": "[user] Help me refactor...",
        "message_count": 3,
        "duration_ms": 4000,
        "ttft_ms": 180,
        "prompt_tokens": 1200,
        "completion_tokens": 800,
        "total_tokens": 2000
      }
    ],
    "errors": [],
    "human_feedback": [],
    "memory_events": [
      {
        "id": "mem-1",
        "phase": "extraction",
        "status": "completed",
        "timestamp": 1720000028.0,
        "title": "Working Memory Extract",
        "summary": "Learned user prefers small focused modules",
        "target_kind": "memory",
        "target_id": "mem-target-1",
        "influence_count": 2
      }
    ],
    "total_events": 12,
    "total_tokens": 2000
  }
}

Key Features

Zero-storage replay: Traces are reconstructed on-demand from the append-only event log — no extra database or video files needed
Seven event types: tool_start, tool_end, llm_call, human_feedback, memory, error, message
Frontend Session Replay Player: The WebUI provides a three-pane interactive player (Chat View / Mind View / Inspector) with scrubber, playback speed control, keyboard navigation, and jump-to-error
Dataset export: Traces can be batch-exported to ShareGPT/Alpaca/OpenAI JSONL format for fine-tuning via the dataset_export pipeline

Growth Statistics

Growth dashboard metrics are available through the unified Learning Journey page (/journey, former /growth auto-redirects) and provide:

Skill KPI summary (total, success rate, evolution count)
Smart savings summary (cost reduction through caching and routing)
84-day activity heatmaps and multi-dimensional health radar charts
Weekly trend analysis and AI-powered daily wrap summaries
Evolution timeline with skill lifecycle events
Knowledge Graph — interactive Claim/Evidence 2D force-directed visualization with namespace filtering
Skill Usage Efficiency Trends — per-skill success rate, average duration, and call frequency over time (daily granularity)

Harness Observability Metrics (Prometheus)

The Myrm Engine exposes underlying Prometheus telemetry metrics for deep DevOps & SRE monitoring. These metrics run on the /metrics endpoint and incur zero operational overhead.

Advanced Agent Metrics

myrm_time_to_first_action_seconds (Histogram): Captures the precise Time-To-First-Action (TTFA), calculating the exact time from receiving a user’s instruction until the agent invokes its first tool.
myrm_policy_denial_total (Counter): Tracks the number of times agent actions were blocked, redacted, or denied by the Security Guardrails and Path Policies.
myrm_tool_execution_total & myrm_tool_execution_failed_total (Counters): Monitors execution results for all individual skills, providing real-time data for calculating the overall “Tool Effectiveness Ratio” on Grafana dashboards.

Production Alerting Rules (Cloud Deployment)

For cloud-hosted deployments, Myrm ships with 12 production-grade Prometheus alerting rules covering sandbox health and reliability:

Alert	Severity	Trigger
ContainerPoolExhausted	Critical	Pool available = 0 for 5 min
ContainerPoolLow	Warning	Pool available < 2 for 10 min
HealthCheckFailureRateHigh	Critical	Failure rate > 20% for 5 min
ContainerCreationSlow	Warning	P99 creation > 10s for 5 min
ContainerOOMKills	Critical	Any OOM kill detected
HotPoolExhausted	Critical	Hot pool = 0 for 5 min

Each alert includes a runbook_url linking to the Enterprise Runbook for remediation steps.

Grafana Dashboard

A ready-to-import Grafana dashboard (grafana-dashboard-sandbox.json) provides real-time visualization of container pool status, hit rates, health check success rates, creation latency, and hot pool availability. Import via Grafana UI or provisioning directory mount.

Smart Enable/Disable

Metrics collection is automatically enabled or disabled based on deployment mode — local and desktop deployments run with zero overhead, while cloud deployments expose the full /metrics endpoint for Prometheus scraping.

Agent Liveness SSOT

A single aggregated endpoint that tells you whether your agent is busy, idle, or degraded — no need to poll multiple APIs.

GET /api/v1/health/liveness

Response

{
  "state": "busy",
  "agents": {
    "activeCount": 2,
    "maxConcurrent": 3,
    "availableSlots": 1,
    "sessions": [
      {
        "sessionId": "abc-123",
        "chatId": "chat-456",
        "agentType": "general",
        "elapsedSeconds": 12.5
      }
    ]
  },
  "channels": {
    "wechat": { "status": "connected" },
    "telegram": { "status": "connected" }
  },
  "memory": {
    "level": "NORMAL",
    "percent": 42.3
  },
  "uptimeSeconds": 3600.5
}

State Logic

State	Condition
`busy`	At least one agent session is actively running
`degraded`	No active sessions, but a channel is disconnected or memory pressure is elevated
`idle`	No active sessions, all channels healthy, memory normal

Use Cases

Frontend tray / pet indicator: Poll this endpoint to show a tri-state icon (spinning = busy, green = idle, amber = degraded)
Cloud monitoring: curl /api/v1/health/liveness | jq .state integrates with Prometheus/Grafana alerting
Multi-pane workspace: Show agent availability across browser tabs without opening settings

Key Properties

Pure read-only aggregation: Zero I/O, zero database queries — all data comes from in-memory gateway state
Sub-millisecond response time: Suitable for frequent polling (every 2-5 seconds)
Works across all deployment modes: Local WebUI, Tauri desktop, and cloud-hosted

​Statistics API

​Daily Journal

​Parameters

​Response

​Data Sources

​Error Responses

​Daily Wrap (AI Summary)

​Parameters

​Response

​Regenerate

​Requirements

​How It Works

​Per-Agent Usage Analytics

​Parameters

​Response

​Key Features

​Session Execution Trace

​Response

​Key Features

​Growth Statistics

​Harness Observability Metrics (Prometheus)

​Advanced Agent Metrics

​Production Alerting Rules (Cloud Deployment)

​Grafana Dashboard

​Smart Enable/Disable

​Agent Liveness SSOT

​Response

​State Logic

​Use Cases

​Key Properties

Statistics API

Daily Journal

Parameters

Response

Data Sources

Error Responses

Daily Wrap (AI Summary)

Parameters

Response

Regenerate

Requirements

How It Works

Per-Agent Usage Analytics

Parameters

Response

Key Features

Session Execution Trace

Response

Key Features

Growth Statistics

Harness Observability Metrics (Prometheus)

Advanced Agent Metrics

Production Alerting Rules (Cloud Deployment)

Grafana Dashboard

Smart Enable/Disable

Agent Liveness SSOT

Response

State Logic

Use Cases

Key Properties