Ready to deploy
your first agent?
On your infrastructure. Your data never leaves. No vendor lock-in. Open source.
Build autonomous agents that plan before they act, SSH into servers, query databases, manage OpenStack clouds, and delegate to each other. Deep planning loop. Dual-path OCR/RAG architecture with self-healing embeddings (Ollama auto-detect, exponential-backoff retries, ChromaDB → Postgres vector fallback). 6-stage RAG pipeline with semantic chunking, hybrid search (RRF), cross-encoder re-ranking, and Redis-backed query cache. On-spot sandbox skills from chat + an Agent Academy to grow your fleet. On your hardware. Your data never leaves.
git clone -b allinone https://github.com/muhammedali275/AI-Orchestrator-Studio && cd AI-Orchestrator-Studio && sudo bash allinone/setup.sh
Every feature you need to go from "idea" to "autonomous agent running in production" — in one platform.
Visual agent creation with custom system prompts, skill assignments, and multi-LLM routing. No code required.
10+ providers (OpenAI, Azure, Anthropic, Ollama, vLLM, Cohere, HuggingFace, LlamaCpp, TextGen, Custom). Per-agent task routing (code_gen → Codellama, rag_answer → GPT-4) with automatic fallback chains.
Team Leader agents delegate to specialists automatically. Recursive multi-agent execution with depth guards.
Semantic chunking → ingest-time embedding → hybrid BM25 + vector search with RRF fusion → metadata filtering → cross-encoder re-ranking → Redis semantic cache.
Auto-detects local Ollama, retries transient failures with exponential backoff, and falls back ChromaDB → Postgres-backed vector store on RHEL/sqlite-old systems. Bulk reindex + embed-pending endpoints survive 1000+ document corpora.
Path A: sync app-server (PyPDF2/PyMuPDF). Path B: async Celery workers with 3-strategy cascade (Tesseract 5 → easyocr → LLM Vision). 8-step image preprocessing. Auto-promote into RAG vector store.
SSH keys, DB logins, API tokens — bind multiple credentials to one agent. Auto-inject by type into matching skills.
AES-256 vault, RBAC, AD/LDAP login, audit trails, TLS, data governance with PII masking and ISO/NIST/GDPR regulatory references.
Agents plan before they act. 4-tier tool enforcement: native function calls → ReAct text parsing → false-completion re-prompt → intent auto-dispatch. 10-round execution loop with self-correction. Works with ANY LLM provider.
15 built-in skills for Nova, Neutron, Cinder, Glance, Keystone, Heat, Swift. Manage compute, network, and storage with natural language.
Cron-based scheduling for automated checks, daily reports, compliance scans. Full execution history tracking.
MS Teams, Slack, Telegram, REST API, Webhooks. Built-in API Gateway with rate limiting and usage analytics.
Redis-backed cache that detects near-duplicate queries via embedding cosine similarity (>0.95). Sub-millisecond cache hits, 24h TTL, per-agent scoping.
Edit .env config from the UI (App, DB, Worker, Storage). Live health-check panel tests PostgreSQL, Redis, ChromaDB, Celery, and vLLM connectivity.
Create & attach a Python skill to an agent in one API call — sandbox enforced by default. Inline Python skills run in a hardened executor with import allow-list, no-network, CPU + wall-clock limits, and stdout capture.
A guided learning track for new agents: import a .skill pack, auto-ingest its references into RAG, run a self-quiz, and graduate to production. Includes ready-made tracks for Presales, Finance, Infra, Legal, and Tibco-style integration agents.
A 3D operations console driven only by real telemetry. If a value can't be measured, it shows — with the reason. It never invents a number.
Agents, LLM connections, skills, connectors and channels are drawn from your real configuration. Edges light up only when real spans flow over them.
A WebSocket stream of spans, metrics, health and alerts from the LLM client, agent executor, skill runner and Celery workers. Under heavy load it switches to batched edge updates.
Click any span to see the full trace waterfall. Click any edge to see call counts, p50 latency, errors and recent calls.
Search agents, skills, LLMs, connectors, documents, conversations and pages, scoped to your tenant.
Tenant admins upload a single logo right on the stage. Files are type-checked, size-limited, virus-scanned and sanitised. Every tenant sees only its own logo.
From user message to tool execution — every layer is deterministic, observable, and provider-agnostic. No black boxes.
Complete message flow from user to final answer — 10 stages, 4 fallback tiers.
Per-agent task-aware model selection with automatic fallback chains. No single-LLM lock-in.
Keyword scoring classifies each message into a task type before selecting the LLM.
Each agent defines its own routing config with primary, task-specific, and fallback LLM connections.
If primary LLM fails (timeout, rate limit, error), the chain automatically tries the next provider.
End-to-end document intelligence — from upload to grounded answer. Self-healing embeddings, vector-store fallback, bulk reindex APIs.
Separated app server, database layer, async workers, and LLM nodes. Designed for horizontal scaling.
React 18 + FastAPI + nginx. Enterprise Orchestrator, UniversalAgentExecutor, ModelRouter, 6-stage RAG, LLMClient.
All relational data, chat memory, rate limiting, task brokering, semantic query cache, vector embeddings.
Async Celery workers. 3-strategy cascade: Tesseract → easyocr → LLM Vision. 8-step image preprocessing. Auto-promotes to RAG.
10+ providers via auto-detection. Per-agent task routing. Primary + fallback chains with cost-aware selection.
From SSH commands to Oracle DBA, from OpenStack cloud to a 6-stage RAG pipeline — agents come ready to work.
Import-ready .skill packages with embedded knowledge bases. Upload → agents start working immediately.
RFP responses, proposals, competitive analysis, objection handling (LAER), ROI/TCO calculations, demo preparation.
Financial analysis, budgeting, forecasting, compliance reporting, revenue recognition, audit preparation.
Server management, monitoring, incident response, capacity planning, patching, backup/recovery procedures.
Contract review, NDA analysis, regulatory compliance, risk assessment, legal terminology, policy drafting.
See how teams across different departments leverage AOS to automate complex, multi-step operations.
Team Leader agent delegates to Linux, VMware, Oracle, and AD sub-agents — all with isolated credentials.
6-stage RAG pipeline: semantic chunking, ingest-time embedding, hybrid RRF search, metadata filtering, cross-encoder re-ranking, and Redis semantic cache.
AD/LDAP enterprise login, regulatory-grade data governance, InfoSec scanning, and full audit trails.
Manage OpenStack/HCS clouds, VMware vCenter, and Kubernetes with natural language. Deep Agent plans multi-step operations before executing.
On your infrastructure. Your data never leaves. No vendor lock-in. Open source.