ContextView

ContextView is the observability and admin hub for ContextUnity — a Django-based dashboard providing trace inspection, agent management, memory browsing, and system analytics.

Trace Observability

Search and browse agent traces, step details, and Langfuse cross-referencing.

Agent Management

CRUD for agent configurations, permissions, tool policies, and hierarchy.

Memory Browser

Inspect episodic and entity memory across tenants and sessions.

MCP Server

7 FastMCP tools for AI-driven admin queries and management.

Architecture

view

Dashboard Views

View	URL	Description
Dashboard	`/`	System overview with KPI charts
Traces	`/traces/`	Agent execution trace list
Trace Detail	`/traces/<id>/`	Step-by-step trace inspection
Agents	`/agents/`	Agent configuration list
Agent Detail	`/agents/<id>/`	Agent config and permissions
Memory	`/memory/`	Episodic and entity memory browser
Events	`/events/`	System event log

gRPC AdminService (20 RPCs)

All RPCs are protected by ViewPermissionInterceptor with two-layer enforcement:

Interceptor — verifies admin:read or admin:write permission
Handler — calls authorize() for defense-in-depth

Category	RPCs	Permission
Agents	ListAgents, GetAgentConfig, GetAgentActivity	admin:read
Agents	UpdateAgentPermissions, UpdateAgentTools	admin:write
Traces	GetTraceDetails, SearchTraces, GetTraceChainOfThought, GetTraceLangfuse	admin:read
Memory	GetMemoryStats, GetMemoryLayerStats, BrowseMemory	admin:read
Health	GetServiceHealth, GetServiceMetrics, CheckServiceConnectivity	admin:read
Analytics	GetSystemAnalytics, GetErrorAnalytics	admin:read
Healing	DetectSystemErrors, TriggerSelfHealing	admin:write
Healing	GetHealingStatus	admin:read

MCP Server (7 Tools)

FastMCP tools for AI-driven admin queries:

Tool	Description
`get_recent_traces`	Recent agent execution traces
`get_trace_stats`	Trace statistics per tenant
`get_episode_stats`	Episode/memory statistics
`get_recent_episodes`	Recent memory episodes
`list_agents`	List all agent configurations
`get_agent_detail`	Get agent config by ID
`brain_search`	Semantic search in Brain

# stdio mode (Cursor/Claude Desktop)
uv run python -m contextunity.view.mcp.server

# SSE mode (web clients)
uv run python -m contextunity.view.mcp.server --sse --port 8101

Data Flow

┌──────────┐     gRPC      ┌──────────────┐     gRPC     ┌────────────┐
│ MCP Tool │ ──────────────▶│ AdminService │ ─────────────▶│   Brain    │
│ (FastMCP)│                │  (Django DB  │               │ (Postgres) │
│          │◀──────────────│   + Brain)   │◀─────────────│            │
└──────────┘  ContextUnit   └──────────────┘  ContextUnit  └────────────┘

┌──────────┐     HTTP       ┌──────────────┐
│ Browser  │ ──────────────▶│  Dashboard   │ (same data sources)
│          │◀──────────────│   (Django)   │
└──────────┘                └──────────────┘

Agent configs → Django DB (local to View)
Traces, episodes, facts → Brain via BrainDataClient
Health checks → direct gRPC to each service
Langfuse URLs → constructed from LANGFUSE_HOST + LANGFUSE_PROJECT_ID

Additional Capabilities

Healing & Analytics — error pattern detection and automated service recovery
Langfuse Integration — cross-references traces between ContextView and Langfuse for deep LLM debugging
Agent Hierarchy — visual permission tree showing agent relationships and capability inheritance