Skip to main content

Dashboard

The WASP web dashboard provides a browser-based interface for chatting with the agent, managing goals and agents, browsing memory, monitoring health, and configuring integrations. It is served on port 8080 and proxied by nginx.

Access​

The dashboard is available at https://your-domain.com and requires authentication.

Login: Use the password set in DASHBOARD_SECRET (minimum 16 characters).

Sessions are persistent (24-hour TTL by default). CSRF protection is enforced on all state-changing endpoints.

The sidebar is organized into 5 labeled sections:

SectionPages
DashboardOverview
ConfigurationsModels, Skills, Skill Evolution, Identity, Integrations, API Keys
SystemConfig, Scheduler, Agents, Goals, Tasks, Memory, Vector Memory, World Model, Knowledge Graph, Cognitive
GovernanceBehavioral Rules, Self-Improve, Subscriptions, Reset
ObservabilityHealth, Audit Log, Live Feed, Metrics

Pages Reference​

Dashboard​

Overview (/)​

The main landing page:

  • Agent identity card (name, personality, born-at, total XP)
  • Current agent state badge (IDLE / ACTIVE / DREAMING)
  • Active goals count, recent skill calls
  • Quick links to all sections

Configurations​

Models (/models)​

Manage LLM providers:

  • All 11 provider cards: Anthropic, OpenAI, Google, xAI, Mistral, DeepSeek, Moonshot/Kimi, OpenRouter, Perplexity, HuggingFace, LMStudio
  • Full model catalog per provider (v2.5: removed 3-model truncation limit)
  • Active provider indicator, health status
  • Switch active model via API

Skills (/skills)​

Manage built-in and custom skills:

  • List all 37+ skills with capability level and description
  • Enable/disable custom Python skills
  • Delete custom skills
  • View skill usage statistics

Skill Evolution (/skill-evolution)​

Automatic skill synthesis management:

  • Detected patterns table (skill combinations + occurrence count)
  • Synthesized skills with diff view
  • Enable/disable synthesis per pattern

Identity (/identity)​

Configure who the agent is:

  • Name, personality description, tone, role definition
  • These fields feed directly into the system prompt (below prime.md)

Integrations (/integrations)​

Configure third-party connectors:

  • List all 40+ available connectors
  • Configure credentials (stored in SecretVault)
  • Set autonomy policy per integration (assist/semi/full)
  • View circuit breaker status (CLOSED / OPEN / HALF_OPEN)
  • Test connector connectivity

API Keys (/apis)​

Manage LLM provider API keys:

  • Set/update keys per provider
  • Keys stored in Redis hash apikeys
  • Masked display for security

System​

Config Center (/config)​

Two-column layout for runtime configuration:

  • Left column: prime.md system prompt editor with Ctrl+S save. Changes take effect immediately (no restart needed).
  • Right column: 12 feature flags organized in 4 groups:
    • Autonomy: SOVEREIGN_MODE, AUTONOMOUS_GOALS_ENABLED
    • Execution: PLAN_CRITIC_ENABLED, META_AGENT_ENABLED
    • Learning: SKILL_EVOLUTION_ENABLED, BEHAVIORAL_LEARNING_ENABLED
    • Memory: VECTOR_MEMORY_ENABLED, WORLD_MODEL_ENABLED, TEMPORAL_REASONING_ENABLED
    • Safety: SELF_IMPROVE_GATE_ENABLED
  • Runtime parameters display (max rounds, timeouts, budget limits)
  • Overrides stored in Redis key config:overrides — survive container restarts

Scheduler (/scheduler)​

Monitor background jobs:

  • All 41 registered jobs with next run time and interval
  • Last run time and duration
  • Success/failure counts per job
  • Manual trigger for any job

Agents (/agents)​

Multi-agent management:

  • List all active and archived agents
  • Create new agents with objectives
  • Pause, resume, and archive agents
  • View current goal per agent
  • Monitor agent resource usage

Goals (/goals)​

Manage autonomous goal execution:

  • List all goals with status, progress, and priority
  • Create new goals with objective and priority
  • Cancel or replan active goals
  • View task graph and execution history per goal
  • Set autonomy mode (full/semi/assist)

Tasks (/tasks)​

User-defined scheduled tasks:

  • List recurring tasks with schedule and last run
  • Create/edit/delete tasks
  • Toggle active/inactive

Memory (/memory)​

Browse the agent's memory:

  • Search episodic and semantic memories by keyword
  • View memory type (episodic, semantic, procedural, visual)
  • Filter by date range and chat ID
  • Delete individual memories
  • Procedural memory tab: view extracted procedures with trigger keywords
  • Dream log tab: view past dream sessions

Vector Memory (/vector-memory)​

Semantic similarity search:

  • Search by natural language query (cosine similarity)
  • View embedding metadata
  • Requires VECTOR_MEMORY_ENABLED=true

World Model (/world-model)​

Real-world entity tracking:

  • Entity state cards (crypto prices, trends, change %)
  • Historical timeline per entity
  • Temporal insights: 24h trend summaries
  • Predictions: LLM-generated forecasts

Knowledge Graph (/knowledge-graph) (extracted from Memory in v2.5)​

Force-directed canvas graph:

  • Physics simulation: node repulsion, link attraction, center pull, drag, click-to-select
  • Node type colors: person (blue), concept (purple), preference (amber), fact (green), organization (cyan), asset (orange), place (lime), event (pink)
  • Node detail overlay panel: label, type, value, confidence, relations
  • Node management table with per-row delete

Cognitive (/cognitive)​

Agent's internal mental state:

  • Self-Model tab: strengths, known failures, user preferences, weekly stats, skill success rates
  • Epistemic tab: domain confidence bars, reset button
  • CPI tab: Cognitive Pressure Index live gauge (0–100), breakdown by dimension, auto-refresh every 30s

Governance​

Behavioral Rules (/behavioral-rules) (new in v2.5)​

Learned rules from user corrections:

  • Stats cards: total / active / disabled / learning examples
  • Type breakdown badges: refusal / hallucination / wrong_skill / missing_context
  • Per-rule: confidence bar, times applied, toggle ON/OFF, delete
  • Client-side text search filter
  • Learning examples table: trigger text, expected skill, use count

Self-Improve (/self-improve) (new in v2.5)​

AI-generated code change proposals awaiting human review:

  • Pending proposals: target file, description, unified diff in <details> viewer
  • Apply (writes file, creates backup, removes from Redis) / Reject (removes from Redis) — both with htmx fade-out
  • Applied history table from AuditLog
  • Proposals stored in Redis hash self_improve:proposals

Subscriptions (/subscriptions) (new in v2.5)​

RSS feeds and price alert management:

  • Stats cards: active, RSS count, price alert count
  • Price alert cards with asset name, threshold, direction (above/below), last check
  • RSS subscriptions table with Pause/Resume/Delete per row

Reset (/reset) (new in v2.6)​

Hard reset the agent to a clean slate:

  • Must type "RESET WASP" exactly in the confirmation field (paste/drop blocked)
  • Progress console streams each reset step live
  • Wipes: 17 DB tables, 12+ Redis key patterns, agent identity XP, self-model
  • Runs VACUUM FULL for immediate disk reclamation
  • AuditLog entry written with action="agent.reset"
  • Result shown in green-bordered card + WaspToast notification

What gets erased: All episodic/semantic/procedural/visual/vector memories, behavioral rules, learning examples, knowledge graph, world timeline, dream log, skill patterns, agents, tasks, capabilities, opportunities, Redis cognitive state.

What survives reset: /data/src_patches backups, API keys, custom Python skill code files, subscriptions, Docker volumes.

Observability​

Health (/health)​

System health dashboard:

  • Redis, PostgreSQL, Ollama connectivity
  • Disk/RAM/CPU gauges with sparklines
  • Health score (0–100)
  • Safety & Execution Control panel (4 sub-panels):
    • Last Soft Gate: last self_improve decision (ALLOW / WARN / BLOCK)
    • CRIT-3 Daily Budget: critical-path self-edits today vs daily cap
    • Saccadic Vision: last content change detection timestamp and hash
    • Learning Queue: behavioral:pending depth / 50 cap with color-coded threshold (yellow ≥20, red ≥40)
  • Integrity Report tab: Self-Integrity Monitor findings — self-model vs actual skill rates, epistemic drift, error spikes
  • Live refresh API: GET /health/api/latest

Audit Log (/audit)​

Paginated audit trail:

  • All actions with timestamp, action type, input/output summaries (secrets redacted)
  • Filter by date range and action type
  • Keyset pagination (cursor-based — O(1) performance)
  • Action types include: skill.shell, skill.self_improve, agent.reset, plus all goal/task actions

Live Feed (/live)​

Real-time event stream:

  • All Redis stream events displayed as they arrive
  • Filter by event type
  • Auto-reconnects on disconnect

Metrics (/metrics)​

Performance charts:

  • Request latency distribution
  • Skill success rates over time
  • Token usage by provider
  • Error rate trends

Streaming Chat​

The chat endpoint uses Server-Sent Events for real-time streaming:

const response = await fetch('/chat/stream', {
    method: 'POST',
    headers: {'Content-Type': 'application/json', 'X-CSRF-Token': csrfToken},
    body: JSON.stringify({message: userInput}),
});

Authentication​

  • Session-based authentication with itsdangerous signed cookies
  • CSRF tokens bound to session ID (not just generated randomly)
  • Login rate limiting (5 attempts/minute)
  • DASHBOARD_SECRET must be at least 16 characters (enforced at startup)

Running Locally​

# Access dashboard directly (bypasses nginx)
ssh -L 8080:localhost:8080 your-server
# Then open http://localhost:8080 in browser

Decision Layer API​

The dashboard exposes a classification endpoint for testing routing:

curl -X POST https://your-domain.com/chat/classify \
  -H "Content-Type: application/json" \
  -d '{"message": "remind me tomorrow at 9am to check email"}'
# Returns: {"strategy": "SCHEDULED_TASK", "confidence": 0.95, ...}