Dashboard

The dashboard is served by agent-core on port 8080. Authentication uses session cookies; the admin user is configured via DASHBOARD_USER / DASHBOARD_PASSWORD. For public-facing deployments, put your own reverse proxy (nginx, Caddy, Cloudflare Tunnel) in front of port 8080 for TLS.

151 HTTP endpoints across the dashboard, organized into five sidebar sections plus a handful of utility surfaces.

Authentication

Sign in at /login. Sessions are signed by DASHBOARD_SECRET (the installer generates 64 random hex chars). Logout at /auth/logout.

CSRF tokens are session-bound — the validator compares the stored token with the session ID, not just the token value. Argon2 password hashing; login rate-limited to 5 attempts per 5 minutes.

If DASHBOARD_USER / DASHBOARD_PASSWORD are not set on first boot, a temporary password is generated and printed once to stderr — wasp logs | grep dashboard.credentials to retrieve it.

Sections

Dashboard

Page	URL	Purpose
Overview	/overview	Top-level snapshot — active goals, recent traces, queue depths
Chat	/chat	Interactive chat (SSE streaming via POST /chat/stream)
Cmd	/cmd	Command palette — search across pages

Configurations

Page	URL	Purpose
Identity	/identity	Agent name, born_at, total_xp, level, persona
Config Center	/config	prime.md editor + 12 feature flags + runtime parameters
Skills	/skills	Installed skills, capability levels, custom skill files
Models	/models	Providers, default model, full catalog (no health-gating)
Integrations	/integrations	40+ connectors with circuit-breaker state

System

Page	URL	Purpose
Scheduler	/scheduler	Recurring tasks, custom tasks, monitor list
Agents	/agents	Sub-agent CRUD, status, message log
Goals	/goals	Active and historical goals with TaskGraph visualization
Subscriptions	/subscriptions	RSS feeds and price alerts
Health	/health	CPI, memory, queues, model liveness

Governance

Page	URL	Purpose
Self-Improve	/self-improve	Pending AI proposals — diff viewer, Apply/Reject, applied history
Behavioral Rules	/behavioral-rules	Learned rules — view, filter, toggle, delete
Audit	/audit	Per-action AuditLog with keyset pagination
Reset	/reset	Panic Reset (hard confirmation gate)

Observability

Page	URL	Purpose
Traces	/traces	Per-response forensic trace
Live	/live	SSE feed of real-time events
Metrics	/metrics	Latency histograms, error rates, token usage
Cognitive	/cognitive	Self-model, epistemic state, integrity report
Memory	/memory	Memory tree browser
Knowledge Graph	/knowledge-graph	Force-directed canvas of entities and relations
World Model	/world-model	Temporal observations, entity timelines
Vector Memory	/vector-memory	Semantic similarity search
Brain	/brain	Stats and persona overview
Tasks	/tasks	Task execution log
Skill Evolution	/skill-evolution	Composite skill candidates + capability evolution
Opportunities	/opportunities	Detected automation opportunities
State	/state	Current execution state

Streaming chat (SSE)

POST /chat/stream opens a Server-Sent Events stream. Each event is a JSON object:

event: progress
data: {"step": 1, "skill": "web_search", "status": "running"}

event: chunk
data: {"text": "Found 12 results. Reading top 3..."}

event: done
data: {"final": "..."}

The chat UI (/chat) consumes the stream and renders progress + final text incrementally.

Decision Trace forensics

/traces is the dashboard's most valuable forensic page. Every response — fast-path, Decision Layer route, or full LLM loop — emits a trace. The page lists traces with filters (chat_id, time range, request tier) and clicking a row shows the full guard chain output.

See Audit Logs → Decision Trace for the schema.

Self-Improve workflow

1. The agent (or you, manually) proposes a patch via self_improve(action="propose").
2. The proposal lands at /self-improve with a diff viewer.
3. You review the diff. If acceptable, click Apply:
   • ast.parse() validates Python syntax.
   • Soft Safety Gate checks for critical-path weakening.
   • Timestamped backup is created at /data/src_patches/backup_*.
   • File is written.
   • The change persists across rebuilds via apply_persisted_patches() at startup.
4. Or click Reject — proposal is dismissed.
5. After Apply, rebuild and recreate agent-core for the change to take effect.

Reset workflow

/reset requires hard confirmation:

• Operator types RESET WASP exactly into a readonly input (paste blocked at the DOM level).
• Submit → progress console streams each step:
  1. Truncate 17 cognitive tables.
  2. Wipe 12+ Redis key patterns.
  3. Reset agent identity (born_at → now, total_xp → 0).
  4. Reset self-model to empty {}.
  5. Run VACUUM FULL.
  6. Write AuditLog entry.
• On completion: green-bordered result card + WaspToast notification.

What survives: API keys, custom Python skills, /data/src_patches/ backups, prime.md, subscriptions.

Health page

The single most important page for daily operation. See Monitoring → /health page.

See also

• Telegram — alternative interface
• Audit Logs
• Configuration
• Monitoring