███████╗███████╗██╗ ██╗███████╗
╚══███╔╝██╔════╝██║ ██║██╔════╝
███╔╝ █████╗ ██║ ██║███████╗
███╔╝ ██╔══╝ ██║ ██║╚════██║
███████╗███████╗╚██████╔╝███████║
╚══════╝╚══════╝ ╚═════╝ ╚══════╝
╰─ by Gyrus Inc ─╯
An open-source, self-hosted agentic framework that turns plain English into SQL Server operations.
Accelerated by the Confluent Accelerator Program for AI Startups (Cohort-3).
Zeus is designed as a professional control plane, incorporating architectural standards validated by the Confluent AI ecosystem.
Need help deploying Zeus or want to see the latest SQL Server agent recipes?
Join the Gyrus Discord Community to connect with other database architects and the Gyrus team.
Quick Start · Features · Architecture · Setup · Safety · Contributing · Discord · Get in Touch
If you find Zeus useful, please consider giving it a star - it helps others discover the project!
Zeus is a self-hosted, multi-agent SQL Server control plane built by Gyrus Inc that lets you manage your SQL Server environment in plain English. Learn more at https://www.thegyrus.com
"create a new schema for finance reporting"
→ Generates and runs the SQL in dependency order
"show me which indexes are missing on my busiest tables"
→ Inspects SQL Server metadata and DMVs, then summarizes the findings
"why are queries blocking in production?"
→ Pulls live operational data and explains the likely cause
Unlike SaaS copilots, Zeus runs in your environment, uses your model provider, and can be customized for your standards and guardrails.
Building and governing SQL Server infrastructure usually means switching between manual SQL, admin tools, scripts, and tribal knowledge. Zeus brings that into one governed workflow.
Beyond object creation, Zeus helps across the full lifecycle:
- Database engineering (schemas, tables, views, procedures, triggers, indexes, jobs) - so delivery teams can move faster with repeatable SQL Server patterns
- Administration (logins, users, roles, grants, database access) - so access and platform setup stay consistent
- Operations and monitoring (DMVs, blocking, sessions, connections, performance checks) - so troubleshooting is faster and more explainable
- Security workflows (row-level security, column-level permissions, policy-oriented access controls) - so sensitive data access can be governed in a structured way
All from natural language, in minutes.
| Self-hosted | Zeus runs in your environment. Credentials stay local. Every line of logic is readable and modifiable. |
| Bring your own model | Works with OpenAI, Anthropic Claude, and Google Gemini out of the box. Swap providers with configuration instead of rewiring the system. |
| Purpose-built for SQL Server | Zeus is built around SQL Server objects, DMVs, permissions, administration workflows, and operational inspection. |
| Safe by design | Destructive SQL is restricted, execution is centralized, and the manager validates outcomes before continuing. |
| Inspection-first | Zeus can inspect live infrastructure before planning changes, reducing blind execution and duplicate work. |
| Natural language all the way | Create objects, inspect metadata, investigate operational issues, and review access patterns from plain English. |
Want to see it in action? Schedule a demo →
git clone https://github.com/Gyrus-Dev/Zeus.git
cd zeus
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txtCopy .env.example to .env, fill in your SQL Server credentials and model API key (see Configure), then:
python -m src.sql_server_ai.objagents.mainFull setup details - model providers, SQL Server connection settings, and runtime options - are in Setup below.
Zeus enforces two independent safeguards before any query reaches SQL Server.
Every agent prefers additive or minimally destructive SQL patterns where possible. Agents are instructed to inspect before acting, execute one step at a time, validate each outcome, and avoid unsafe SQL unless a specific workflow explicitly requires it.
A centralized execution tool intercepts every statement before it reaches SQL Server.
DROP- blocked outright.TRUNCATE- requires explicit approval.- Execution flow - results are recorded in state so the manager can validate before the next step.
User request
|
v
Agent generates SQL
|
v execute_query safety gate
| |- contains "DROP"? -> blocked
| |- contains "TRUNCATE"? -> requires approval
| `- otherwise -> passed through
|
v
execute_query() -> SQL Server
Because Layer 2 is enforced in code, not only in prompts, it gives Zeus a more reliable control point than direct model-to-SQL execution.
┌──────────────────────────────────────────────────────────────┐
│ CLI (Rich + prompt_toolkit) │
└──────────────────────────────┬───────────────────────────────┘
│ user message
v
┌──────────────────────────────────────────────────────────────┐
│ SQLSERVER_ARCHITECT (Manager) │
│ Classifies intent · plans work · delegates one step at │
│ a time · validates results through state │
└──────┬──────────┬──────────┬────────────┬────────────┬──────┘
v v v v v
DATA ADMINISTRATOR SECURITY INSPECTOR ACCOUNT
ENGINEER ENGINEER PILLAR MONITOR
\___________ _________ _________ _________ _________/
|
v
execute_query() -> SQL Server
|
v
session state validation
| Pillar | Role |
|---|---|
| SQLSERVER_ARCHITECT | Manager - plans, routes, validates |
| DATA_ENGINEER | Databases, schemas, tables, indexes, views, procedures, functions, triggers, SQL Server Agent jobs |
| ADMINISTRATOR | Logins, users, roles, grants, access configuration |
| SECURITY_ENGINEER | Row-level security and column-level permission patterns |
| INSPECTOR_PILLAR | Read-only inspection of SQL Server metadata and catalog state |
| ACCOUNT_MONITOR | Query stats, blocking, sessions, connections, and operational health |
| RESEARCH_AGENT | Documentation and best-practice fallback |
- You type a natural language request.
- The Manager classifies intent and creates an execution plan.
- Pillar agents receive delegated tasks one at a time.
- Specialist agents generate and execute SQL Server statements through the shared execution tool.
- After every step, the manager validates success from session state before continuing.
- CLI output shows execution progress and tool activity.
- On exit, Zeus can preserve the executed query trail for later review.
Use plain English to perform common SQL Server administration workflows.
"create a login and user for the analytics team"
"set up a new schema for finance"
"grant read access to the reporting role"
Zeus routes these requests through SQL Server-specific agents that understand object dependencies, access patterns, and execution order.
Zeus can inspect SQL Server infrastructure before taking action.
"show me all tables in the sales schema"
"list stored procedures in reporting"
"check whether this index already exists"
This supports discovery-first operations and reduces duplicate or unsafe changes.
Zeus can answer day-2 operational questions using SQL Server metadata and DMV-style inspection.
"why is the database slow right now?"
"show blocking sessions"
"which queries are consuming the most resources?"
This makes Zeus useful not just for build workflows, but also for operations and incident investigation.
Pro Tip: Encountering a complex blocking issue or performance bottleneck? Describe the scenario in the
#help-and-supportchannel on our Discord, and we'll help you
craft the perfect agent sequence to diagnose it.
Zeus helps with governed access setup and security-related SQL Server patterns.
Examples include:
- users and logins
- roles and grants
- row-level security patterns
- column-level permission handling
- schema-scoped access configuration
Zeus supports multiple model-provider configurations.
- Google Gemini
- Anthropic via LiteLLM
- OpenAI via LiteLLM
That means the orchestration layer can remain stable even if your model strategy changes.
Data Engineering
Databases · Schemas · Tables · Indexes · Views · Stored Procedures · Functions · Triggers · SQL Server Agent Jobs
Administration
Logins · Users · Roles · Grants · Database access configuration
Security
Row-level security workflows · Column-level permission patterns
Inspection and Monitoring
System catalog inspection · Information schema discovery · Query stats · Sessions · Connections · Blocking and operational checks
Zeus is open-source and community-driven. We aren't just building a tool; we're creating a hub for Agentic Data Ops across all major platforms. Join our Discord to:
- 🛠️ Get Agent Recipes: Discover the most effective prompt sequences for automating SQL Server migrations and performance tuning.
- 🗺️ Influence the Roadmap: Vote on new features and tell us which SQL Server operational patterns we should automate next.
- ⚡ Direct Access: Get real-time architectural guidance and support from the creators at Gyrus Inc.
- 🧪 Beta Access: Get early access to new specialist agents and experimental pillars before they hit the main branch.
## Setup
### Prerequisites
- Python 3.11.10
- A reachable SQL Server instance
- ODBC Driver 18 for SQL Server
- An API key for your chosen model provider
### Install
```bash
git clone https://github.com/Gyrus-Dev/Zeus.git
cd zeus
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
Create a .env file in the project root by copying the provided template:
cp .env.example .envThen fill in your values - refer to .env.example for all available variables and their descriptions.
| Variable | Required | Description |
|---|---|---|
SQLSERVER_HOST |
Yes | SQL Server hostname or IP |
SQLSERVER_PORT |
Yes | SQL Server port |
SQLSERVER_USER |
Yes | SQL Server username |
SQLSERVER_PASSWORD |
Yes | SQL Server password |
SQLSERVER_DATABASE |
No | Default database |
SQLSERVER_DRIVER |
No | ODBC driver name |
SQLSERVER_TRUST_SERVER_CERTIFICATE |
No | Trust server certificate for local/dev environments |
| Variable | Required | Description |
|---|---|---|
APP_USER_NAME |
Yes | Display name shown in session state |
APP_USER_ID |
Yes | Unique user ID for session tracking |
APP_NAME |
Yes | Application name for session scoping |
| Variable | Required | Description |
|---|---|---|
MODEL_PROVIDER |
No | google (default) · openai · anthropic |
GOOGLE_API_KEY |
If google |
API key for Gemini models |
OPENAI_API_KEY |
If openai |
API key for OpenAI models |
ANTHROPIC_API_KEY |
If anthropic |
API key for Claude models |
MODEL_PRIMARY |
No | Override the primary model |
MODEL_THINKING |
No | Override the reasoning model |
- terminal-first interactive experience
- execution progress visibility
- debug mode for agent payloads and generated code
- centralized SQL execution feedback
zeus/
├── src/sql_server_ai/
│ ├── objagents/
│ ├── adksession.py
│ ├── adkstate.py
│ └── adkrunner.py
├── skills/
├── .env.example
├── README.md
└── requirements.txt
- Python 3.11
- Google ADK
- LiteLLM
- Rich
- prompt_toolkit
- pyodbc / SQL Server ODBC driver
- Join the Discord community: https://discord.gg/fErydWMv
Contributions, bug reports, and documentation improvements are welcome.
Zeus is designed to be adaptable. You can customize prompts, skills, guardrails, and routing behavior to match your SQL Server standards and operating model.
If you want Zeus adapted for your environment, internal standards, or multi-system workflows, reach out to Gyrus.
- Website: https://www.thegyrus.com
- Email: priyank@thegyrus.com
- Discord: https://discord.gg/fErydWMv
This project is licensed under the Apache 2.0 License. See LICENSE.