Step by Step

From zero to autonomous AI team in minutes

Sutra is designed to be the simplest way to get AI agents working for you. Here's exactly how it works — from installation to your first automated workflow.

Before You Start

What you'll need

Three things. That's it. If you have a modern computer, you're probably already set.

💻

A Computer

Mac, Windows, or Linux. Sutra runs on all three. You'll need at least 4 GB of free RAM.

Required

🐳

Docker Desktop

Docker handles all the setup for you — the database, background services, everything. One install and you're done.

Required

🔑

An AI API Key

From OpenAI, Anthropic, Google, or others. Or skip this entirely and use Ollama to run free AI models locally.

Optional with Ollama

Step 1 · 2 Minutes

Install with a single command

Clone the project and run the install script. It pulls everything you need, sets up Docker containers, and creates your configuration file. No manual setup required.


                        # Clone the project

                        $ git clone https://github.com/datar-gaurav/sutra-os.git

                        $ cd sutra-os


                        # Run the installer

                        $ ./install.sh


                        # That's it. Open your browser.

⚙

Automatic Configuration

The install script creates your environment file and asks for your API keys. Paste them in and you're configured.

📦

Everything Included

Database, cache, background workers, API server, and dashboard — all start automatically inside Docker.

Step 2 · 2 Minutes

Set up your AI API keys

Head to the Settings page and add your API keys. Sutra supports every major provider — OpenAI, Anthropic, Google Gemini, Groq, and more. Or skip this entirely and use Ollama to run powerful AI models for free, locally on your machine.

🔑

All Major Providers

Add keys from OpenAI, Anthropic, Google, Groq, Mistral, or any OpenAI-compatible endpoint. Switch between them any time.

💻

Run Locally with Ollama

No API key? No problem. Connect Ollama and run Llama, Mistral, Qwen, or any open model free on your own hardware.

🛡

Keys Stay On Your Machine

Your API keys are stored locally in your environment file and never sent to any third-party server — only directly to the AI provider.

💰

Budget Controls

Set monthly spend limits per provider. Sutra tracks token usage in real-time and alerts you before you hit your cap.

Step 3 · 2 Minutes

Create a Purpose — define your model strategy

A Purpose is a named routing profile that defines which AI models to use and in what priority order. Instead of hard-coding a model into every agent, you assign a Purpose. When the primary model is unavailable or overloaded, Sutra automatically falls back to the next one — zero downtime, zero manual intervention.

🧭

Named Profiles

Create purposes like “All Purpose”, “Coding”, or “Local First”. Each one bundles a list of models ranked by preference.

🔄

Automatic Fallback

If P1 is down or rate-limited, Sutra silently tries P2, then P3. Your agents keep working without you lifting a finger.

🎯

Task-Optimised Routing

Use a fast, cheap model for simple tasks and a powerful one for complex reasoning — just by picking the right Purpose for each agent.

💡

Mix Any Provider

Combine cloud APIs and local Ollama models in the same priority list. Run local-first for privacy, fall back to the cloud when needed.

⚡

One change, every agent. Update a Purpose and every agent assigned to it instantly picks up the new model strategy — no need to edit each agent individually.

Step 4 · 3 Minutes

Create your first AI agent

Open the dashboard and create an agent. Give it a name, a role, and tell it what it should do. Assign a Purpose for its model strategy, pick which tools it can use, and it's ready to work.

🎭

Choose a Role

Content writer, research analyst, customer support, project manager — or make up your own. You define what the agent does.

🧰

Assign Tools

Give your agent access to Gmail, Slack, web search, GitHub, spreadsheets, or any of 30+ built-in integrations.

📋

Use Templates

Not sure where to start? Pick from ready-made agent templates that cover the most common use cases.

🧠

Assign a Purpose

Select the Purpose you created and the agent inherits the full model priority strategy — with automatic fallback built in.

Step 5 · When You're Ready

Build automated workflows

Once you're comfortable with individual agents, connect them into workflows. The visual builder lets you design multi-step automations by dragging and dropping — no code needed.

🎨

Visual Canvas

Drag nodes onto a canvas and draw connections between them. Each node is a step: an AI call, a decision point, a tool action, or a human approval.

🔀

If/Then Logic

Route work based on conditions. "If the sentiment is negative, escalate to a human. If positive, send a thank-you email."

🔄

Loops

Process a list of items automatically. Upload 50 resumes and have an agent screen each one, scoring and summarizing the results.

✋

Human Checkpoints

Insert approval gates at any point. The workflow pauses, you review and approve, then it continues. You're always in control.

Step 6 · Scale Up

Let your agents work as a team

Create multiple agents and let them collaborate. Set up structured discussions where agents brainstorm, debate, or review each other's work before delivering a final result.

💬

5 Discussion Formats

Brainstorm, debate, standup, review, and collaborate. Each format structures how agents communicate to get the best results.

🏢

Org Structure

Organize agents into teams with managers and reports. Work flows through the hierarchy naturally, just like a real organization.

Step 7 · Ongoing

Monitor everything from one dashboard

The dashboard gives you a complete view of what your agents are doing. Track conversations, review pending approvals, monitor costs, and see full audit logs of every action taken.

📊

38+ Dashboard Pages

Agent performance, workflow status, conversation history, budget tracking, and system health — all in one place.

🛡

Approval Queue

High-risk actions go to a review queue. Approve or reject with one click. Nothing sensitive happens without your say-so.

💰

Cost Tracking

See exactly how much each agent and workflow costs in API usage. Set budgets and get alerts before limits are reached.

📝

Full Audit Trail

Every decision, tool call, and output is logged. Review the complete history of what happened and why.

Use Cases

See it in action

Here are a few real examples of what people build with Sutra — from simple tasks to complex multi-agent workflows.

📧

Automated Email Triage

An agent monitors your inbox, categorizes incoming emails, drafts replies for routine questions, and flags urgent items for your attention.

→ Email arrives in Gmail
→ Agent reads and classifies it
→ Routine: drafts reply for your approval
→ Urgent: sends you a Slack alert

📝

Content Pipeline

A research agent gathers information, a writer agent creates the draft, and a reviewer agent checks for quality — all automatically.

→ You provide a topic
→ Researcher finds sources and data
→ Writer creates the article draft
→ Reviewer suggests improvements

📋

Customer Feedback Analysis

An agent pulls reviews from multiple sources, identifies common themes, and creates a weekly summary with action items.

→ Scrapes reviews from web sources
→ Analyzes sentiment and themes
→ Creates summary in Google Sheets
→ Posts highlights to Slack

💻

Code Review Assistant

Forge reviews pull requests on GitHub, checks for common issues, suggests improvements, and can even write the fix itself.

→ New PR is opened on GitHub
→ Agent reviews code changes
→ Posts review comments
→ Optionally writes fix PRs

📊

Weekly Report Generator

At the end of each week, an agent compiles data from your tools, writes a summary, and emails it to your team.

→ Pulls data from Sheets and Jira
→ Summarizes progress and blockers
→ Generates formatted report
→ Emails to the team via Gmail

🔍

Competitive Intelligence

Agents continuously monitor competitor websites, news, and social media, alerting you when something important changes.

→ Scheduled web scraping of competitors
→ AI compares changes to last scan
→ Flags significant updates
→ Delivers briefing via Slack or email

FAQ

Common questions

Quick answers to the things people ask most.

No. Sutra is designed so that anyone can create agents, build workflows, and manage automations through a visual interface. You'll only need to use the terminal for the initial installation (copy-paste three commands), and after that everything is point-and-click.

Yes. Sutra is 100% open-source and free to use. The only cost is the AI model API usage (e.g., OpenAI charges per request). You can eliminate even that cost by using Ollama to run free, local AI models on your own machine.

It depends on which AI model you use and how much you use it. GPT-4o costs roughly $2.50–$10 per 1M tokens. For most users, that's a few dollars per month. Sutra's built-in budget controls let you set limits so you never get a surprise bill. Or use Ollama for $0.

Yes. Upload documents to the built-in knowledge base and your agents will automatically search them when answering questions. Since Sutra runs on your own machine, your data never leaves your control.

Sutra has multiple safety layers. You can classify actions by risk level, require human approval for sensitive operations, set budget limits, and review a complete audit trail of every action taken. You're always in the loop.

Yes. Deploy Sutra on a shared server and your team members can access the dashboard from their browsers. Each person can create agents, run workflows, and manage their own automations from the same instance.

Ready?

Three commands. Five minutes.
Your AI team is running.

Clone the repo, run the installer, and open your browser. Everything else is handled for you.

Clone on GitHub Explore All Features