Rig Agent Runtime Platform¶

Rig Agent Runtime is the shared agent runtime. Same Docker image, different character.json per agent. All rig agents run on Rig Agent Runtime.

Repo: Stig-Johnny/rig-agent-runtime (existing, to be evolved)

Three-Layer Architecture¶

┌──────────────────────────────────────────────────────────────┐
│ Layer 1: Rig Agent Runtime Runtime (shared image, Node.js)          │
│                                                              │
│  Messaging Adapters    Agent Loop    LLM Gateway             │
│  ┌──────────┐         ┌─────────┐   ┌──────────────┐        │
│  │ Discord  │◀───┐    │ Process │   │  OpenClaw    │        │
│  │ Adapter  │    │    │ message │──▶│  (all models │        │
│  └──────────┘    │    │ → tools │   │   Claude,    │        │
│  ┌──────────┐    ├───▶│ → reply │   │   GPT, etc.) │        │
│  │ Slack    │    │    └─────────┘   └──────────────┘        │
│  │ Adapter  │◀───┘                                          │
│  └──────────┘         ┌─────────┐   ┌──────────────┐        │
│  ┌──────────┐         │ Memory  │   │  Dashboard   │        │
│  │ Future   │         │ (Postgres│   │  (WebSocket) │        │
│  │ (Teams)  │         │  + Redis)│   │              │        │
│  └──────────┘         └─────────┘   └──────────────┘        │
│                                                              │
│  Webhooks    CronJobs    Health    Tool Executor             │
└──────────────────────────────────────────────────────────────┘

┌──────────────────────────────────────────────────────────────┐
│ Layer 2: Agent Config (per agent, no code)                   │
│                                                              │
│  character.json    values.yaml    tools config               │
│  (personality,     (k8s resources, (HTTP endpoints,          │
│   platform,         secrets,       MCP servers)              │
│   model,            replicas)                                │
│   channels)                                                  │
└──────────────────────────────────────────────────────────────┘

┌──────────────────────────────────────────────────────────────┐
│ Layer 3: Backend APIs (per agent, when needed)               │
│                                                              │
│  rig-conductor API (.NET 10)     AI Accountant API           │
│  - Marten event store          - Accounting integration      │
│  - Assignment engine           - Receipt processing          │
│  - Escalation logic            - Fiken/Folio API             │
│  - Human gate checker                                        │
│  - Priority queue                                            │
└──────────────────────────────────────────────────────────────┘

Current State → Target State¶

Feature	Current (v0.1)	Target
Messaging	Discord only (`discord.js`)	Adapter interface: Discord, Slack
LLM	`@anthropic-ai/sdk` direct	OpenClaw gateway (all models)
Config	Discord-specific `character.json`	Platform-agnostic config
Production	rig-agent-runtime on GCP k3s (was Book-E pre-2026-05-20)	rig-agent-runtime hosting all rig agent personalities

Platform Evolution¶

Messaging Adapter Interface¶

Current Rig Agent Runtime has Discord hardcoded throughout. Target: abstract behind an adapter interface.

interface MessagingAdapter {
    connect(): Promise<void>
    sendMessage(channel: string, content: string): Promise<void>
    onMessage(handler: (msg: Message) => void): void
    createThread(channel: string, name: string): Promise<string>
    addReaction(channel: string, messageId: string, emoji: string): Promise<void>
}

class DiscordAdapter implements MessagingAdapter { ... }
class SlackAdapter implements MessagingAdapter { ... }

Configured in character.json:

{
  "messaging": {
    "platform": "discord",
    "config": {
      "guildId": "1477260986854498477",
      "channels": {
        "tasks": "1477261755125075969",
        "admin": "1477263870845517879"
      }
    }
  }
}

Swap to Slack:

{
  "messaging": {
    "platform": "slack",
    "config": {
      "workspace": "dashecorp",
      "channels": {
        "tasks": "C0123TASKS",
        "admin": "C0123ADMIN"
      }
    }
  }
}

LLM via OpenClaw¶

Current: direct @anthropic-ai/sdk calls. Target: route through OpenClaw.

{
  "llm": {
    "provider": "openclaw",
    "endpoint": "http://openclaw.gateway:8080/v1",
    "model": "claude-opus-4-6",
    "fallbackModel": "claude-sonnet-4-6",
    "temperature": 0.3,
    "maxTokens": 8192
  }
}

Benefits:

Model flexibility — switch models per agent without code changes
Cost tracking — OpenClaw tracks usage per agent
Routing — OpenClaw can route to cheapest/fastest model per task
Fallback — automatic fallback on model unavailability
All providers — Claude, GPT, Gemini, local models — whatever OpenClaw supports

character.json Full Schema¶

{
  "name": "rig-conductor",
  "description": "Engineering rig coordinator",
  "personality": "Precise, efficient coordinator. Speaks in short, factual messages.",

  "messaging": {
    "platform": "discord | slack",
    "config": { }
  },

  "llm": {
    "provider": "openclaw | anthropic | openai",
    "endpoint": "http://...",
    "model": "claude-opus-4-6",
    "temperature": 0.3,
    "maxTokens": 8192
  },

  "memory": {
    "conversationRetention": "30d",
    "patternRetention": "indefinite",
    "historyRetention": "5y"
  },

  "tools": [
    {
      "type": "http",
      "name": "event-store",
      "description": "Submit and query events from the engineering rig event store",
      "baseUrl": "http://rig-conductor-api:8080",
      "endpoints": [
        { "method": "POST", "path": "/api/events", "description": "Submit an event" },
        { "method": "GET", "path": "/api/queue", "description": "Get priority queue" },
        { "method": "GET", "path": "/api/agents", "description": "Get agent status" }
      ]
    },
    {
      "type": "mcp",
      "name": "github",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"]
    }
  ],

  "cron": [
    {
      "schedule": "*/1 * * * *",
      "prompt": "Run the coordination loop: check milestones, agent health, stuck agents, human gates, assign work."
    }
  ],

  "webhooks": [
    {
      "path": "/github",
      "events": ["issues", "pull_request", "check_run"],
      "prompt": "Process this GitHub event: {{payload}}"
    }
  ]
}

How Each Agent Uses Rig Agent Runtime¶

Agent	character.json	Backend API	Notes
rig-conductor	Coordinator personality, cron every 60s, GitHub webhook, event store tools	.NET API (Marten event store, assignment, escalation)	The brain
Dev-E	Developer personality, Claude Code CLI tool, GitHub tools	None — runs Claude Code directly	The hands
Book-E (retired 2026-05-20)	Accountant personality (succeeded by AI Accountant)	AI Accountant API	Replaced by AI Accountant service on the cluster
Review-E	Reviewer personality, Claude Code CLI, GitHub tools	None — runs Claude Code directly	Migrated into the cluster (rig-agent-runtime pod, GCP k3s) 2026-05-20; formerly Pi4-02
Monitor-E	Monitor personality, health check tools, alerting	Health check API (Phase 3)	Phase 3
Architect-E	Architect personality, analytics tools	Analytics API (Phase 4)	Phase 4

Dashboard¶

Rig Agent Runtime's built-in dashboard provides per-agent:

Live log stream (WebSocket)
Active sessions
Tool call history
Token usage + cost tracking
Conversation threads

rig-conductor's backend API adds rig-specific views:

Priority queue (kanban)
Agent assignment status
Escalation panel
Human gate queue
Milestone progress
Event store browser

Both dashboards accessible at rig.dashecorp.com (Cloudflare Access).

Deployment¶

All agents deploy as Helm releases on the same Rig Agent Runtime chart:

# rig-gitops/apps/rig-conductor/helmrelease.yaml
apiVersion: helm.toolkit.fluxcd.io/v2
kind: HelmRelease
metadata:
  name: rig-conductor
  namespace: rig
spec:
  chart:
    spec:
      chart: charts/rig-agent-runtime
      sourceRef:
        kind: GitRepository
        name: rig-agent-runtime
  values:
    character: rig-conductor
    image:
      repository: ghcr.io/stig-johnny/rig-agent-runtime
      tag: latest
    resources:
      requests: { cpu: 100m, memory: 256Mi }
      limits: { cpu: 500m, memory: 512Mi }
    persistence:
      enabled: true
      size: 1Gi

# rig-gitops/apps/dev-e/helmrelease.yaml
apiVersion: helm.toolkit.fluxcd.io/v2
kind: HelmRelease
metadata:
  name: dev-e
  namespace: rig
spec:
  chart:
    spec:
      chart: charts/rig-agent-runtime
      sourceRef:
        kind: GitRepository
        name: rig-agent-runtime
  values:
    character: dev-e
    image:
      repository: ghcr.io/stig-johnny/rig-agent-runtime
      tag: latest
    resources:
      requests: { cpu: 500m, memory: 512Mi }
      limits: { cpu: "2", memory: 4Gi }