Skip to main content

voro-core

The operational backbone. Runtime infrastructure for the OpenClaw agent execution environment: gateway configuration, audited execution wrappers, knowledge collections, and workspace templates.

Purpose

voro-core provides the secure execution environment for autonomous agents. It manages runtime mode selection, audited command execution with resource limits, model proxy configuration, network allowlisting, and the Mission Control event ingestion API.

Architecture

Runtime Modes

ModeEnvironmentDescription
local-rigWSL2 / LocalDefault development mode
vps-prodDigitalOcean VPSHardened production deployment

Mode is controlled by infra/mode.env and infra/lib/mode.sh.

Core Infrastructure

ComponentPurpose
run_with_audit.shAudited command wrapper with time/memory limits, before/after snapshots, allowlist enforcement
openclaw.jsonGateway + model + agent sandbox configuration template
network_allowlist.txtOutbound destination whitelist for network-isolated execution
docker-compose.ymlLiteLLM model proxy plane (loopback-only)
litellm_config.yamlModel routing: Anthropic Sonnet 4.6, Haiku 4.5

Mission Control (Event Ingestion)

The Mission Control server is the operational backbone — an HTTP event ingestion layer for fleet-wide agent telemetry.

RouteMethodPurpose
/GETEmbedded HTML dashboard
/api/healthGETService health
/api/ingestPOSTEvent ingestion (single or array, schema v2.0)
/api/agentsGETFleet status list
/api/eventsGETEvent query with filters
/api/costsGETCost rollup query
/api/events/streamGETSSE event stream (real-time)

Rate limits: Root 60rpm, API 240rpm, Stream 60rpm.

Event Schema (v2.0)

Required fields: schema_version, agent_id, agent_role, timestamp (ISO 8601), event_type, severity, message

Supported event types: agent lifecycle (agent.start/stop/heartbeat), LLM operations (llm.request), tool usage (tool.*), predictions, trading, scanning, policy violations.

Module Inventory

ModuleResponsibility
apps/mission_control/HTTP event ingestion (server.py 863L), SQLite persistence, SSE streaming, fleet status, cost rollups
infra/Audited execution, LiteLLM proxy, deployment modes, network allowlisting
tools/~50 Python scripts — MC smoke tests, policy validators, pilot lifecycle, source management
knowledge/collections/5 agent memory collections (trading, coding, cloud)
WORKSPACE_TEMPLATE/Agent persona definitions (AGENTS.md, SOUL.md), behavioral contracts, skill specs

Database

Runtime DB: data/mission-control.db (SQLite, WAL mode)

SchemaTablesPurpose
Contract v2events, cost_rollups, agent_status, ingest_errorsCore event store, fleet health, cost tracking
Contract v3pilot_cohorts, recommendation_ledger, cohort_resultsPilot cohort management, weekly gate recommendations
Contract v3.1pilot_phases, pilot_audit_logLifecycle state transitions, governance audit trail

Key Capabilities

  • Audited execution: Every command runs through run_with_audit.sh with time/memory limits and before/after snapshots
  • Network isolation: Outbound connections restricted to explicit allowlist
  • Model proxy: LiteLLM routes model requests through a loopback-only proxy, preventing direct API exposure
  • Event ingestion: Fleet-wide telemetry collection with structured event schema
  • SSE streaming: Real-time event stream for dashboard consumption
  • Workspace templates: Agent persona definitions with behavioral contracts and skill specifications
  • Knowledge collections: Curated agent memory search paths for trading, coding, and cloud operations

Build & Run

# Start Mission Control
python3 -m apps.mission_control.server --host 127.0.0.1 --port 3900 --token "$MC_TOKEN"

# Validate gateway config
python3 -c "import json; json.load(open('infra/openclaw.json.example'))"
bash infra/run_with_audit.sh openclaw gateway probe

# Health check
curl -sS http://127.0.0.1:3900/api/health | jq .

# Smoke tests
python3 tools/mission_control_smoke.py
python3 tools/mission_control_api_smoke.py

Current State

  • Language: Python 3.12
  • Tests: 197 pytest passed
  • Database: SQLite WAL mode (~148KB)
  • Shipped work: MC-001 through MC-053 (complete)
  • Status: Stable. P3/P4 capability expansion drafts await operator sign-off.