Release Notes

A summary of features, improvements, and fixes shipped in each release of Menatic AI.

v1.0.0 — February 12, 2026

Initial public release of the Menatic AI platform.

Local models — deploy and manage Ollama models directly on the cluster; pull new models from the UI
Remote models — connect OpenAI, Google Gemini, or any OpenAI-compatible inference provider
Per-model MCP server assignment (controls which tools a model can call)
GPU resource configuration for the Ollama StatefulSet
OpenAI-compatible API (/api/v1/chat/completions, /api/v1/models) for SDK compatibility

One-click JupyterLab deployment as Kubernetes Deployments
Resource presets: Standard (1 CPU / 2Gi), Medium (2 CPU / 4Gi), Large (4 CPU / 8Gi), and fully custom
Optional GPU allocation per notebook
Token-secured auto-login URLs — no manual token entry
Pre-installed openai, requests, ipywidgets, and papermill libraries
Pre-injected environment variables: FTN_BASE_URL, FTN_API_KEY, FTN_PROJECT_ID
Example notebook pre-seeded showing OpenAI SDK usage against the platform API

Submit any notebook from the shared workspace as an unattended Kubernetes Job via Papermill
Parameterised execution — inject key-value pairs into tagged notebook cells at runtime
Resubmit jobs with one click
Per-job logs streamed in the UI
Output notebooks written to shared/batch_outputs/<jobID>/ and browsable from JupyterLab
Shared workspace PVC (10Gi RWX) auto-provisioned and mounted across all notebooks in a project

Generate a fully deployed MCP server from any OpenAPI specification using LLM-assisted code generation
Four-stage async pipeline: code generation → Kaniko image build → container registry push → Kubernetes deployment
Real-time generation logs in the UI
Per-server health status monitoring
Admin Chat integration: create and list MCP servers via natural language

Upload PDF and plain-text documents per project
Automatic text extraction, chunking (1 800-character chunks, 200-character overlap), and embedding via nomic-embed-text
Per-project vector collections stored in Qdrant with cosine similarity search
Semantic search API (/api/rag/search) returning scored, ranked chunks
Full document lifecycle: upload (multipart or JSON), list, inspect, delete
Delete removes both metadata and all associated vectors
Usage stats endpoint aggregating document and chunk counts per project

Three-level hierarchy: Organization → Department → Project
Each project gets a dedicated Kubernetes namespace with enforced ResourceQuota (CPU, memory, storage, GPU)
Per-project API tokens for programmatic access from notebooks and external services
Per-project MCP server allow-list
Cascading deletes across the hierarchy

Token consumption tracking on every chat completion (prompt, completion, total tokens)
Tracked per model, provider, and project
Time-series report with daily/hourly buckets (24h, 7d, 30d, 90d presets)
Hierarchical report with org → department → project drilldown

Multi-turn AI assistant powered by a configurable platform model
Full suite of cluster management tools: get cluster status, list pods/deployments, configure Ollama, pull models, create notebooks and MCP servers
MCP protocol (SSE + JSON-RPC) for tool invocation