Roadmap¶
Last verified: v2.0
Current State: v2.0¶
MEHO v2.0 represents a comprehensive platform with cross-system diagnostic intelligence, 15 connector types, and a production-ready deployment model.
What's Shipped¶
MEHO has gone through extensive development across 10 milestones, delivering:
- Cross-system reasoning -- trace problems across Kubernetes, cloud infrastructure, observability stacks, CI/CD pipelines, and collaboration tools in a single conversation
- 15 connector types -- Infrastructure (Kubernetes, VMware, Proxmox, GCP), Observability (Prometheus, Loki, Tempo, Alertmanager), CI/CD (ArgoCD, GitHub), Collaboration (Jira, Confluence, Email), and Generic (REST via OpenAPI, SOAP via WSDL)
- Trust model -- three-tier classification (READ/WRITE/DESTRUCTIVE) with approval workflows and audit trail for every operation
- Intelligent data pipeline -- JSONFlux engine with Arrow tables, Parquet caching, and DuckDB SQL reduction that keeps LLM context manageable even with large datasets
- Topology auto-discovery -- automatic entity extraction and cross-system resolution (providerID, IP, hostname matching) that builds a connected infrastructure graph
- Dual-mode chat -- Ask mode for knowledge base Q&A, Agent mode for cross-system investigation
- Three-tier knowledge architecture -- global, connector-type, and connector-instance scoping with hybrid search (BM25 + semantic) and Voyage AI reranking
- Token optimization -- 81% cumulative reduction through observation compression, sliding windows, stateful loop management, and step budgets
- Security hardening -- Content Security Policy, HSTS, CORS lockdown, memory-only auth tokens, Keycloak OIDC integration
- Investigation visualization -- hypothesis tracking, citations, breadcrumb navigation, and topology animation showing investigation paths
v2.0 Stabilization¶
The current stabilization milestone focused on hardening what's built:
- Repository cleanup -- archived planning artifacts, resolved all ESLint and Ruff warnings, updated pre-commit hooks
- Testing infrastructure -- layered test conftest, PydanticAI test model integration, verified test suites
- Bug fixes -- resolved critical and major bugs, implemented AWS/Azure providerID parsing, removed dead code and stale references
- Documentation -- complete documentation rewrite covering all features, all connectors, and deployment guides
What's Next¶
Testing Expansion¶
Expanding automated test coverage beyond the current baseline:
- End-to-end test suite -- automated browser tests covering all user journeys (currently 5 Playwright specs, expanding to full coverage)
- CI pipeline -- GitHub Actions with automated test gates, coverage thresholds, and deployment checks
- LLM evaluation -- automated assessment of agent response quality using LLM-as-judge evaluation pipelines
- Performance baselines -- regression testing for critical paths (agent response time, data pipeline throughput)
Internal Documentation¶
Engineering reference material for the development team:
- Architecture decision records -- extracting and documenting the key decisions made during development
- Connector development guide -- step-by-step guide for adding new connector types to the platform
- Agent prompt engineering guide -- best practices for authoring skills and tuning agent behavior
Performance Optimization¶
Fine-tuning the platform for production workloads:
- Query optimization -- database query analysis and indexing improvements
- Caching strategy -- intelligent cache invalidation and pre-warming for frequently accessed data
- Streaming performance -- SSE connection management and message delivery optimization
New Connector Types¶
Expanding platform coverage based on customer needs:
- Additional cloud providers (AWS, Azure native connectors)
- Database connectors (PostgreSQL, MySQL, MongoDB)
- APM integrations (Datadog, New Relic, Dynatrace)
- Container orchestration (Docker Swarm, Nomad)
Each new connector type follows the established pattern: typed connector class, operation definitions with trust classification, topology entity extraction, and auto-generated skills.
Platform Features¶
Capabilities under consideration for future milestones:
- Investigation replay -- post-investigation timeline scrubber for reviewing past diagnostic sessions
- Scheduled reports -- automated periodic investigations with report delivery
- Team collaboration -- shared investigation sessions with real-time presence
- Custom dashboards -- saved query templates for recurring diagnostic patterns
Contributing¶
MEHO's roadmap is driven by real-world operational needs. If you have suggestions for new connector types, features, or improvements, reach out to the evoila engineering team.