Automate-E Runtime
Automate-E is a Kubernetes-native AI agent runtime for Discord. It turns a character.json config file into a running Discord bot backed by Claude, with persistent memory, tool calling, and cost tracking.
Key Features
| Feature | Description |
|---|---|
| Character-driven | Define personality, tools, and behavior in a single JSON file |
| Claude-powered | Agent loop uses Anthropic Claude API with tool use |
| Postgres memory | Conversations, user facts, and merchant patterns persist across restarts |
| In-memory fallback | Works without Postgres for local development |
| Tool calling | Agents call HTTP APIs defined in character.json |
| Cost tracking | Per-model token usage and cost calculation |
| Live dashboard | Real-time WebSocket dashboard for monitoring agent activity |
| K8s-native | Designed to run as a Deployment with ConfigMap-mounted character files |
How It Works
sequenceDiagram
participant U as Discord User
participant G as Discord Gateway
participant A as Agent Loop
participant C as Claude API
participant T as Tool APIs
participant M as Memory (Postgres)
U->>G: Message in #invoices
G->>A: messageCreate event
A->>M: Load conversation history + user facts
A->>C: System prompt + history + tools
C->>A: tool_use: get_folio_balance
A->>T: GET /folio/balance
T->>A: {balance: 142350}
A->>C: Tool result
C->>A: Text response
A->>M: Save messages
A->>G: Reply in thread
G->>U: "Saldo: 142 350 kr"
Repository
The runtime lives in Stig-Johnny/automate-e (private). Agent configurations (like Book-E) live in the consuming repo alongside their k8s manifests.
Quick Links
- Quick Start -- run an agent locally in 5 minutes
- Configuration -- full
character.jsonreference - Architecture -- how the runtime works internally
- Deployment -- deploy to Kubernetes
- Book-E -- the first Automate-E agent