XYBEROS — AI Platform

A Python framework for building cognitive AI systems with a full pipeline: perceive, reason, plan, decide, act, reflect, and remember.

Quick Start

pip install xyberos

from xyberos import xyberos

ctx = xyberos.chat("Hello, world!")
print(ctx.thought.summary)

Or with Docker:

docker compose up

With observability (OpenTelemetry + Prometheus + structured logging):

pip install xyberos[observability]

From source:

git clone https://github.com/xyberos/xyberos.git
cd xyberos
make install    # pip install -e ".[dev,docs,web]"
make test       # run all tests
make docs       # build documentation

Features

• Cognitive pipeline — 8-stage processing loop with pluggable strategies
• Plugin system — unified Plugin contract with 12 entry-point groups, 54 auto-discovered components
• Security subsystem — authentication, authorization, RBAC, Guardian decision engine, risk scoring, audit trail, capability management, policy engine, human-in-the-loop approval, resource governance
• 6-tier memory — working, episodic, semantic, procedural, vector, knowledge graph with automatic fallback
• Multi-agent runtime — agent communication, shared memory, task delegation, collaborative voting
• Model-agnostic LLM — 9 backends: OpenAI, Anthropic, Ollama, OpenAI-compatible, Google Vertex AI, AWS Bedrock, Azure OpenAI, vLLM, simulated
• Pluggable tools & skills — web search, file I/O, calculator, shell commands, and custom extensions
• REST API — FastAPI server with OpenAPI docs, versioned endpoints (/v1/), SSE streaming, WebSocket support
• Authentication middleware — JWT token creation/validation, API key auth, Bearer token support
• Secrets management — HashiCorp Vault integration with automatic env-var fallback and TTL caching
• Multi-tenant security — Tenant isolation in storage and database via key prefixing and column injection
• Plugin marketplace — Community plugin registry with publishing guide
• Configuration — YAML + env vars + programmatic overrides
• Docker — one-command deployment with docker compose up
• ML evaluation — built-in classification and regression metrics
• OpenTelemetry integration — export traces and metrics to Jaeger, Datadog, Grafana
• Prometheus metrics — kernel-level metrics at /metrics with auto-collected request timing
• Structured logging — JSON-formatted logs ready for Loki / Datadog / Splunk
• Graceful shutdown — KillSwitch wired to SIGINT/SIGTERM for clean teardown
• Runtime introspection API — GET /system/status and GET /system/info endpoints
• Makefile — make test, make lint, make build, make docs, make docker-build

Documentation

Full documentation is available in the docs/ directory:

• Getting Started
• XYBEROS API & Assistant
• Creating Tools
• Multi-Agent Systems
• Memory System
• Security Subsystem
• Configuration
• Plugin Discovery & Providers
• API Reference

Stability

XYBEROS is currently in pre-release (0.x). The public API is stabilizing but may change between minor versions. Breaking changes are documented in CHANGELOG.md.

See VERSIONING.md for the versioning policy.

Requirements

• Python 3.12+

License

Apache 2.0 — see LICENSE.