Agent Command Center
Every panel backed by NucleusDB
A web dashboard with 9 pages covering observability, cost analysis, trust attestation, live terminal cockpit, agent deployment, and database administration. The CRT visual language — scanlines, grain, terminal contrast — is part of the product identity.
agenthalo dashboardNine Pages, Complete Control
#/overviewOverview
System status, recent sessions, cost summary, trust score, quick actions. Your starting point.
#/sessionsSessions
Session list with search and filters. Click into any session for full event-level trace detail with timestamps.
#/costsCosts
Daily cost charts, per-agent breakdown, per-model comparison. Identify spending patterns and optimize model selection.
#/configConfiguration
Shell wrapping setup, x402 payment configuration, vault key management, and general settings.
#/trustTrust
Session attestations with Merkle proofs, trust scores, epistemic calculi metrics, and formal verification status.
#/nucleusdbNucleusDB
SQL console for direct database queries. Vector search interface. Commitment proof verification.
#/cockpitCockpit
Live terminal panels with xterm.js + WebGL rendering. PTY bridge for interactive agent sessions. Diversity gauge and trace topology charts.
#/deployDeploy
Agent catalog with launch cards. Preflight checks, admission policy evaluation, governor snapshot before launch.
#/modelsModels
Local model management for Ollama and vLLM backends. Install, remove, and configure models.
Live Terminal Cockpit
Full terminal emulation in the browser. xterm.js with WebGL rendering and PTY bridge.
Full terminal emulation via PTY manager. WebSocket bridge connects browser xterm.js to server-side pseudo-terminals.
Multiple agent terminals side by side. Monitor several agents running concurrently.
Tsallis diversity metric for strategy diversity monitoring. Real-time gauge showing agent exploration breadth.
H0 persistence (Rips complex) chart showing trace structure. Visualize how agent actions cluster and connect.
Deploy & Launch
Agent Catalog
Pre-configured agent profiles for Claude Code, Codex, Gemini CLI. Custom agent configurations.
Preflight Checks
Validate environment, credentials, model availability, and resource requirements before launch.
Admission Policy
Governor-based admission control. Evaluate compute and PTY governors before allowing new agent sessions.
Launch Controls
Set working directory, model, temperature, and allowed tools. One-click launch with full parameter control.
Multi-Agent Orchestrator
Launch and coordinate multiple agents in parallel with task dependency management.
Agent Pool
Launch and manage multiple agents concurrently. Each agent gets its own PTY session and trace stream.
Task Graph DAG
Define multi-step task dependencies as a directed acyclic graph. Agents execute tasks in topological order.
A2A Mesh Delegation
Delegate tasks across agents via DIDComm HTTP bridge. Cross-agent coordination with typed message passing.
Trace Bridge
Aggregate traces from all agents in the pool into a unified timeline. Cross-agent event correlation.
Container Isolation & Mesh Network
Each agent can run in an isolated Docker container with automatic mesh registration. Cross-container MCP tool calls enable agent cooperation without shared filesystem access.
Container Builder
Generates Dockerfiles from agent configuration. Injects identity, MCP bindings, and observability hooks at build time.
Agent Lock
Filesystem-level coordination primitives. Prevents concurrent container operations from corrupting shared state.
Mesh Registration
Containers auto-register on the libp2p mesh. DHT discovery + Gossipsub for peer awareness.
Sidecar Process
Health monitoring sidecar alongside each agent. Reports liveness, resource usage, and trace statistics.
Cross-Container MCP
Invoke MCP tools across container boundaries via mesh routing. JSON-RPC 2.0 with DIDComm authentication.
Subsidiary Registry
Register and discover subsidiary agent pools. Hierarchical orchestration with parent-child delegation.
API Proxy & Local Models
OpenAI-compatible multi-provider proxy. Routes requests to cloud providers (via OpenRouter) or local backends (Ollama, vLLM) automatically. Token pricing, vault-managed API keys, and HTTP 402 payment support.
OpenRouter API with vault-managed keys
localhost:11434, auto-detected models
localhost:8000, GPU-accelerated inference
Your Agents, Your Dashboard
Everything runs locally. Every panel backed by verifiable storage.