Admin Panel

The Sagewai Admin Panel is a full observability and control suite for managing your AI agent infrastructure. It provides real-time monitoring, cost analytics, compliance dashboards, budget controls, and interactive tools for testing agents and workflows.

Running the Admin Panel

Backend (FastAPI, port 8000):

make dev APP=admin
# or with Docker:
make dev-up  # starts all services including admin at :8000

Frontend (Next.js, port 3008):

cd clients/web && pnpm dev --filter @sagecurator/admin

Open http://localhost:3008 to access the dashboard.

Architecture

The admin panel has three layers:

All data stores are in-memory for the MVP — designed for easy swap to PostgreSQL-backed implementations in production.

Dashboard

URL: / (root page)

The main dashboard provides a high-level overview of your agent infrastructure with 6 KPI stat cards:

• Registered Agents — total agents in the registry
• Total Runs — completed + running agent invocations
• Active Sessions — currently open multi-turn conversations
• Total Tokens — aggregate token usage across all models
• Total Cost — cumulative LLM spend in USD
• Risk Events — PII detections + hallucination flags + content filter triggers

Below the cards:

• Cost Trend Chart — line chart showing cost over time, useful for spotting spend spikes
• Model Usage Chart — pie chart of token distribution by model, showing which providers consume the most resources

API endpoints used: GET /admin/agents, GET /admin/runs, GET /admin/sessions, GET /api/v1/analytics/costs, GET /api/v1/analytics/usage, GET /api/v1/analytics/risks

Agent Management

Agent Registry

URL: /agents

Lists all registered agents in a table with columns:

Agent Detail

URL: /agents/{name}

Detailed view of a single agent showing:

• Configuration: Model, status, max iterations, total runs
• Capabilities: Purple badge tags
• Tools: List of MCP tools available to the agent
• System Prompt: Preview of the agent's system prompt (first 500 chars)

Runtime Config Updates

You can update an agent's configuration at runtime via the API without restarting:

curl -X PATCH http://localhost:8000/admin/agents/my-agent/config \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-4o-mini", "temperature": 0.5, "max_iterations": 15}'

Updatable fields: model, system_prompt, temperature, max_tokens, max_iterations.

Run History

URL: /runs

Browse all agent invocations with filtering:

• Agent Name — text filter to narrow by agent
• Status — dropdown: All, Completed, Running, Failed

Each run shows: Run ID, Agent, Status (colored badge), Token count, Start timestamp.

Run Detail

URL: /runs/{id}

Full detail of a single run including:

• Summary: Agent name, status, total tokens, duration
• Input: The user message that triggered the run
• Output: The agent's final response
• Steps: Each iteration of the agent loop with type, duration, and detail
• Tool Calls: Each tool invocation with name, arguments, result preview, and duration

Run Controls

Active runs can be controlled in real-time:

Live Sessions

URL: /sessions

Monitors all active multi-turn conversation sessions. Each session shows:

• Session ID, Agent name, Status, Message count, Start time

Sessions are created by AdminState.start_session() and removed when end_session() is called.

Observability

Cost Analytics

URL: /analytics/costs

Detailed cost breakdown with:

• Budget Alert Banner — yellow warning when agents are approaching their spend limits
• 4 Stat Cards: Total Cost, API Calls, Models Used, Active Agents
• Cost by Model — horizontal bar chart comparing spend across providers
• Breakdown Tables — two tables (by Model and by Agent) with columns: Name, Cost, Tokens, Requests

API: GET /api/v1/analytics/costs, GET /api/v1/analytics/models, GET /api/v1/analytics/agents

Model Comparison

URL: /analytics/models

Side-by-side comparison of all models in use:

• 4 Stat Cards: Models Active, Total Requests, Cheapest Model, Most Used Model
• Comparison Table with sortable columns:
  • Model name, Provider (auto-inferred), Total Cost, Total Tokens
  • Request Count, Cost per 1K Tokens

Use this to identify which models offer the best cost-performance ratio for your workloads.

Execution Monitor

URL: /monitor

Real-time monitoring of agent executions via Server-Sent Events (SSE):

• Connection Controls: Connect/Disconnect SSE stream, Demo Mode for testing
• Agent Filter: Dropdown to filter events by agent
• Run List Sidebar: Scrollable list of active/recent runs with status dots and event counts
• Event Timeline: Chronological timeline of AG-UI events for the selected run, with timestamps and event type badges

Event types include: run_started, text_message_start, text_message_content, tool_call_start, tool_call_end, state_snapshot, run_finished, run_error.

Compliance

PII Compliance Dashboard

URL: /compliance/pii

Monitors PII detection and redaction activity across all agents:

• 4 Stat Cards: PII Detections, Hallucination Flags, Content Filtered, Total Events
• PII Events Chart — timeline showing detection frequency over time (lines for PII, hallucination, content filter)
• Entity Breakdown — chart showing distribution of detected entity types (EMAIL, PHONE, SSN, CREDIT_CARD, IBAN, IP_ADDRESS, PASSPORT)
• Redaction Actions Table — summary of actions taken (Protected/Warned/Blocked) with counts
• Export Audit Report — button to download a compliance report

API: GET /api/v1/analytics/risks

PII detection is powered by the PIIGuard guardrail. See the PII Protection guide for setup instructions.

Operations

Budget Manager

URL: /operations/budget

Configure and manage per-agent spending limits:

• Budget Limits Table — all configured limits showing: Agent, Daily Limit, Monthly Limit, Action (badge), Fallback Chain, Edit/Delete buttons
• Create/Edit Form:
  • Agent Name
  • Max Daily USD
  • Max Monthly USD
  • Action: Warn (log and continue), Throttle (switch to cheaper model), Stop (halt execution)
  • Fallback Model Chain — ordered list of models to fall back to when budget is exceeded
• Fallback Chain Editor — visual editor with numbered priority, up/down/remove buttons, and model dropdown

Budget Actions

Example: Setting Up Budget with Fallback

1. Navigate to /operations/budget
2. Click "+ New Limit"
3. Set agent name: my-agent, daily limit: $5.00, monthly: $100.00
4. Set action: Throttle
5. Add fallback chain: gpt-4o-mini → gemini-2.0-flash → llama-3.1-8b-instant
6. Save

When my-agent exceeds $5/day, it automatically switches from its primary model to gpt-4o-mini, then gemini-2.0-flash, then llama-3.1-8b-instant.

API endpoints:

See the Cost Management guide for SDK-level integration.

Tools

Playground

URL: /playground

Interactive environment for testing agents without writing code:

• Left Panel — Agent Configuration:
  • Name, Model (dropdown), Strategy (dropdown), System Prompt, Temperature slider (0–2)
  • "Create Agent" button to instantiate an ad-hoc agent
• Right Panel — Chat:
  • Real-time streaming chat via SSE (AG-UI protocol)
  • Messages appear token-by-token as the agent generates them
  • Tool calls are displayed inline as they execute

Available strategies: react, lats, tot, self_correct, planning, routing.

API: POST /playground/agent (create), POST /playground/run (SSE stream)

Strategy Lab

URL: /strategy-lab

Compare different execution strategies side-by-side on the same prompt:

1. Select a model
2. (Optional) Set a system prompt
3. Check the strategies you want to compare (multi-select)
4. Enter your test prompt
5. Click "Run Comparison"

Results appear in a grid, one card per strategy showing:

• Output text
• Duration (ms), Tokens used, Cost (USD)
• Number of reasoning steps and tool calls
• Status badge (success/error)

A summary bar highlights the fastest and cheapest strategy.

API: POST /strategies/compare (SSE stream with comparison_started, strategy_result, comparison_finished events)

Workflow Builder

URL: /workflows

Visual editor for building and testing YAML workflows:

• Left Panel:
  • Template selector — pre-built workflow templates to start from
  • YAML editor — syntax-highlighted textarea for editing workflow definitions
  • Inline validation — green/red indicator showing if the YAML is valid
• Right Panel:
  • Pipeline graph — visual representation of the agent pipeline
  • Test input field and "Run Workflow" button
  • Execution log — dark-themed log showing workflow events in real-time

Templates include common patterns like sequential pipelines, parallel fan-out, and loop-until-done workflows.

API: POST /workflows/validate (check YAML), GET /workflows/templates, POST /workflows/run (SSE stream)

Fleet Management

The fleet pages let you manage distributed workers that execute agent tasks across your infrastructure.

Workers

URL: /fleet/workers

Lists all registered workers with real-time status:

• Status badges — online (green), offline (gray), draining (yellow)
• Pool and label filtering — narrow the list by worker pool or JSONB labels
• Model capabilities — each worker displays its available models (normalized via ModelNormalizer)
• Approval status — pending, approved, or rejected badges reflecting the approval state machine
• Enterprise badges — IP allowlist, dual approval, and WebSocket indicators for enterprise-tier workers

Worker Detail

URL: /fleet/workers/{id}

Detailed view of a single worker with 3 tabs:

• Overview — worker metadata, pool assignment, labels, model capabilities, probe status display (Ollama + OpenAI-compatible health probes), connection history
• Tasks — task execution history for this worker with status and duration
• Anomalies — anomaly alert cards showing rate limit violations, excessive failures, heartbeat timeouts, and model mismatch detections (auto-revoke triggers on 2+ anomaly types)

Enrollment Keys

URL: /fleet/enrollment

Manage enrollment keys that workers use to register with the fleet:

• Create Key — generate a new enrollment key with an optional description and expiry
• Key List — table of all keys with creation date, status, and usage count
• Revoke Key — immediately invalidate a key (revoked keys cannot be reused)

Workers enroll using the CLI (sagewai fleet register) or Docker worker images with the enrollment key passed as an environment variable.

Audit Log

URL: /fleet/audit

Chronological log of all fleet events with filtering:

• Event type filter — 13 event types including worker registration, approval, task dispatch, heartbeat, and anomaly detection
• Severity filter — info, warning, error levels
• Time range — filter events by date range

API endpoints: GET /api/v1/fleet/workers, GET /api/v1/fleet/workers/{id}, POST /api/v1/fleet/enrollment-keys, GET /api/v1/fleet/audit

Context Engine

The context engine pages manage document ingestion, retrieval, and knowledge lifecycle.

Context Overview

URL: /context (also shown as a section on /intelligence/dashboard)

High-level overview of your context data:

• Scope stats grid — document counts, chunk counts, and storage metrics split by scope (org-level vs project-level)
• Ingestion activity — recent ingestion jobs with status indicators

Documents

URL: /context/documents

Full document management interface:

• Multi-file upload — drag-and-drop or file picker with per-file progress bars and tag input on upload
• Directory indexing — index an entire directory via native folder picker or by entering a server path
• Search — text search with debounce across document titles and metadata
• Sortable columns — sort by name, source, scope, chunk count, created date
• Pagination — paginated results for large document collections
• Batch delete — multi-select checkboxes with batch delete action
• Auto-poll — processing status updates automatically while ingestion is in progress

Document Detail

URL: /context/documents/{id}

Detailed view of a single ingested document:

• Metadata — source type, scope, tags, ingestion timestamp, file size
• Chunk list — all chunks extracted from the document with content previews and token counts
• Source info — original file path or URL, content type, language detection results

Search

URL: /context/search

Multi-strategy retrieval interface:

• Strategy selection — vector (semantic), BM25 (keyword), graph (entity-relation), or combined (RRF merge)
• Scope selector — search within org-level, project-level, or both
• Tag filtering — filter results by document tags
• Results display — ranked results with relevance scores, source attribution, and chunk previews

Lifecycle

URL: /context/lifecycle

Configure automatic document lifecycle management:

• Lifecycle triggers — set thresholds for compress, archive, and discard actions based on document age or access frequency
• Maintenance panel — manually trigger lifecycle operations (compression, archival, cleanup)
• Conflict table — view and resolve conflicting facts detected during re-ingestion (auto-supersede or manual review)

Directives

URL: /context/directives

Reference page for the directive engine syntax:

• Directive catalog — all supported directives (@context, @memory, @agent:name(), @wf:name(), /tool, #meta)
• Dynamic parameters — documentation for @datetime, @date, @time, @user, @project with usage examples
• Template syntax — {{ }} template expressions with examples

API endpoints: GET /api/v1/context/documents, POST /api/v1/context/documents, DELETE /api/v1/context/documents, POST /api/v1/context/search, GET /api/v1/context/lifecycle

Memory

Vector Store

URL: /memory/vector

Browse and inspect Milvus vector collections:

• Collection list — all Milvus collections with document count and dimensionality
• Search — run similarity searches against stored embeddings
• Collection stats — storage size, index type, and segment information

Knowledge Graph

URL: /memory/graph

Interactive graph exploration powered by NebulaGraph:

• Entity browser — searchable, paginated list of all entities in the graph
• Entity detail — view an entity's properties, relations, and neighbors
• Interactive navigation — click on related entities to traverse the graph
• Relation viewer — inspect edge properties including temporal facts (valid_from, superseded_at)

API endpoints: GET /api/v1/graph/entities, GET /api/v1/graph/entities/{id}/neighbors, GET /api/v1/graph/entities/{id}/relations

Intelligence

Intelligence Dashboard

URL: /intelligence/dashboard

Overview of the intelligence layer capabilities:

• Feature grid — status of each intelligence component (embedder, language detector, entity extractor, relation extractor, fact extractor, summarizer, transcriber, vision describer)
• Consolidation metrics — semantic dedup count, exponential decay scores, contradiction detection stats
• Provider info — active provider for each component (local, API, or hash fallback)

LLM Spend

URL: /intelligence/spend

Monitor LLM usage costs in real-time:

• Global spend — total cost across all models with trend indicator
• Per-model breakdown — table of each model with request count, token usage, and cost
• Auto-refresh — polling interval that pauses when the browser tab is not visible (visibility-based polling)
• Time range selector — filter spend data by day, week, or month

API endpoints: GET /api/v1/analytics/costs, GET /api/v1/analytics/models

Projects and Workspaces

Project Management

URL: /settings/projects

Manage project-scoped isolation for all data stores:

• Project list — all projects with name, creation date, and agent count
• Create project — set up a new project with name and description
• Switch project — workspace switcher in the sidebar for quick context switching
• Default project — auto-created on first startup via ensure_default_project(); cannot be deleted

Multi-project support is a premium feature. Free-tier accounts operate within a single default project.

Project Settings

Per-project configuration options:

• LiteLLM virtual key — assign a virtual API key scoped to this project for spend tracking and rate limiting
• Budget config — set project-level daily and monthly spend limits with warn/throttle/stop actions
• Notification overrides — per-project notification channel configuration (override global SMTP/Slack/SSE settings)

API endpoints: GET /api/v1/projects, POST /api/v1/projects, PUT /api/v1/projects/{id}

Infrastructure Settings

Infrastructure

URL: /settings/infrastructure

Configure connections to external infrastructure services:

• LLM Proxy — LiteLLM proxy URL and masked API key with a "Test Connection" button to verify connectivity
• Vector DB — Milvus URI configuration for vector storage
• Graph DB — NebulaGraph host and port settings

All connection details are stored encrypted. Changes take effect on the next agent invocation without requiring a restart.

Notifications

URL: /settings/notifications

Configure notification channels for budget alerts, workflow events, and system warnings:

• SMTP Email — server, port, credentials, and sender address for email notifications
• Slack Webhook — webhook URL for posting alerts to a Slack channel
• In-App SSE — server-sent events channel for real-time browser notifications

Each channel can be enabled or disabled independently. Per-project overrides allow different channels for different workspaces.

Connectors

URL: /settings/services

Manage external service integrations:

• Connector list — all 18 builtin connectors with health badges (healthy, degraded, disconnected)
• Search and filter — find connectors by name, category, or health status
• Connector detail — 3 tabs:
  • Overview — connector description, status, last health check timestamp
  • Configuration — credential fields (API keys shown masked), OAuth settings, endpoint URLs
  • Tools — MCP tools exposed by this connector with name and description
• Auto health polling — periodic health checks with automatic badge updates

API endpoints: GET /api/v1/connectors, GET /api/v1/connectors/{id}, GET /api/v1/connectors/{id}/health

Eval

Datasets

URL: /eval/datasets

Upload and manage evaluation datasets:

• Upload — import eval datasets in JSON format with input/expected-output pairs
• Dataset list — all datasets with name, sample count, and creation date

Run Eval

URL: /eval/run

Execute evaluation runs against your agents:

• Model selection — choose which model to evaluate
• Dataset selection — pick one or more datasets to run
• Run execution — start the evaluation and monitor progress in real-time

Reports

URL: /eval/reports

View evaluation results:

• Report list — all completed eval runs with model, dataset, score, and timestamp
• Report detail — per-sample results with input, expected output, actual output, and judge scores

API endpoints: POST /api/v1/eval/run, GET /api/v1/eval/reports

API Reference Summary

All admin endpoints at a glance:

For detailed request/response schemas, see the REST API Reference.