Skip to content

Web UI

Nexus includes a built-in web interface at /ui/ for visual inspection and monitoring of your agents. No separate installation is required — the UI ships with grampus-ai[server] and is served directly from the Grampus FastAPI server.

Starting the UI

pip install "grampus-ai[server]"
grampus serve
# Open http://localhost:8000/ui/

uv run uvicorn grampus.server.app:create_app --factory --host 0.0.0.0 --port 8000
# Open http://localhost:8000/ui/

grampus dev
# Open http://localhost:8000/ui/

The API documentation is available alongside the UI at http://localhost:8000/docs.

Dashboard (/ui/)

The dashboard gives a real-time overview of your agent fleet:

  • Active agents — count of currently running agent sessions
  • LLM calls today — total model calls since midnight UTC
  • Cost today — accumulated USD spend since midnight UTC
  • Errors today — count of failed runs and safety violations

Below the stat cards is the Recent alerts table showing the last 20 alert events — which rule fired, severity, agent, and timestamp.

The dashboard auto-refreshes every 10 seconds via HTMX. No manual page reload needed.

Memory Inspector (/ui/memory/)

The memory inspector lets you browse, search, and manage all memory entries for your agents without writing code.

Filter bar

At the top of the page, filters narrow the displayed entries:

Filter Description
Agent ID Show only records for a specific agent
Memory type episodic, semantic, procedural, or all
Search text Full-text search across record content
Min trust score Hide low-trust entries (e.g., set to 0.5 to see only moderate/high trust)

Hit Apply to reload the table with new filters. Filters are bookmarkable — they update the URL query string.

Memory table

Each row shows:

Column Description
Type Memory type badge: episodic, semantic, or procedural
Content Truncated content preview (click row to expand)
Trust Trust score with color coding
Source SourceType of the provenance record
Created Timestamp
Actions Delete button

Trust score colors

Color Score range Meaning
Green ≥ 0.8 High trust (system or validated user input)
Yellow 0.5–0.8 Moderate trust (LLM-generated or tool results)
Red < 0.5 Low trust (external data or flagged content)

Detail panel

Click any row to slide open the detail panel on the right, showing:

  • Full content (untruncated)
  • Complete Provenance metadata (source type, source ID, content hash)
  • Importance score and access count (episodic)
  • Subject/predicate/object and confidence (semantic)
  • Procedure steps (procedural)

Deleting entries

Click the trash icon in the row Actions column to delete a single entry. You will be prompted to confirm. Deletion is immediate and irreversible — the entry is removed from Dapr state.

Upcoming pages

The Evals and Cost pages are coming in the next release (D10). The sidebar shows them as placeholders. Check the sidebar navigation: the active page is highlighted; placeholder pages are grayed out.

Programmatic access

All data shown in the UI is available via the REST API — useful for automation and integration with external dashboards:

# List memory entries
curl "http://localhost:8000/memory?agent_id=research-bot&memory_type=episodic"

# Delete a memory entry
curl -X DELETE "http://localhost:8000/memory/ep-001"

# Dashboard stats
curl "http://localhost:8000/ui/_stats"

# Recent alerts
curl "http://localhost:8000/alerts/history?limit=20"

Full API documentation is at http://localhost:8000/docs.

See also