Themis
SRE agent platform with hexagonal architecture, finite state machines, Kafka, Qdrant and Prometheus integration. 19 ADRs published.
The problem
SRE scales badly. More services = more alerts = more burnout. Existing AIOps platforms (Moogsoft, BigPanda) are closed-box, expensive, and don’t understand each team’s specific architecture.
What Themis does
Themis is a multi-domain SRE agent platform that adapts to the client’s architecture. Each agent represents a domain (Kafka, Kubernetes, PostgreSQL, Aether) with its own finite state machine and reasoning.
Agents consume telemetry from Kafka/Prometheus, reason with a context-aware LLM via Qdrant embeddings, and produce auditable actions (runbook, Linear ticket, Slack message).
Current state
19 ADRs published. 3 critical bugs identified (BUG-001/002/003) pending fix before implementing the core.
Architecture
- Hexagonal — reasoning domain doesn’t depend on specific Kafka or LLM. Swappable adapters.
- FSM per agent — agent state is readable. No “vibes-based AI”.
- MCP Server included — any Claude/Cursor client can query Themis state via MCP.
Stack
Go 1.25 · franz-go · Qdrant · Prometheus · MCP SDK.
Want to talk about this?
If Themis resonates with a problem you're facing, book a 30-minute call to get into the detail.