Summary

Suna is an open-source general-purpose AI agent: you describe the goal, and it works inside an isolated container using a browser, files, the command line and third-party APIs to get the job done — useful for batch data work, web research and report generation where the task must actually be completed.

Suna: an open-source general AI agent that actually executes tasks screenshot

What Suna is

Suna is Kortix's open-source general-purpose AI agent. It connects a chat interface to a set of tools that can actually do things: you describe the task in plain language, and the agent uses a browser, the file system, a shell and third-party APIs inside an isolated container to carry it out — instead of just returning a paragraph of advice. That execution ability is what matters for work that has to be finished end to end: filling in a spreadsheet in bulk, pulling information from web pages, producing an analysis report.

The code is public on GitHub and the project supports self-hosting, so you can deploy it yourself and bring your own model keys, or use the hosted version.

Core capabilities

  • Browser automation: open pages, paginate and extract data to complete tasks that need multiple clicks.
  • File management: create, edit and manage documents, including text and spreadsheets, and generate report files.
  • Crawling and search: fetch page content, organise the data and produce summary reports.
  • Shell execution: run system commands inside the container for automation scripts and admin tasks.
  • Site deployment: deploy and manage websites, useful for getting an online service up and maintained quickly.
  • API integrations: connect third-party services such as LinkedIn and Crunchbase to pull in and process external data.
  • Data analysis and reporting: turn collected data into market analysis, review summaries or paper comparisons.
  • Live interaction: the run reports back as it goes, so you can see which step it is on and where it stalls.

How it is built

  • Backend: Python with FastAPI, handling REST endpoints, thread management and connections to model providers such as OpenAI and Anthropic.
  • Frontend: Next.js with React, providing the chat interface and task dashboard.
  • Agent Docker: each task runs in its own container, so the browser, code interpreter, file system and tool calls stay isolated.
  • Supabase: handles persistence — authentication, user management, conversation history, file storage, agent state and realtime subscriptions.

Typical usage

State the goal and the constraints clearly in natural language; the agent plans the steps, calls the tools and reports as it works. Three common patterns:

  1. Batch data enrichment: hand it a spreadsheet of B2B leads, ask it to find each company's LinkedIn page, identify the CEO or founder and the head of sales or marketing, add a one-line description of the business, and write everything back into the file.
  2. Targeted research: ask what a product or company is and what it does, and get a structured write-up.
  3. Deep analysis: provide an analysis frame — company overview, financials, market sentiment, technical picture, competitor comparison, valuation and risk — and let it produce the full report.

Pricing

  • Free: 10 minutes of runtime per month.
  • Pro: $29 per month, 4 hours of runtime.
  • Enterprise: $199 per month, 40 hours of runtime.

Billing is by runtime, not by token, and that is the main thing to weigh when estimating cost: the longer an agent runs, the faster the allowance goes.

Where it fits

  • Market and business analysis, competitor and industry-trend reviews.
  • Data collection, such as policy documents, public tenders or social media data.
  • Personal and administrative work: trip planning, report writing, file organisation.
  • Education and research: paper summaries, study-material organisation.
  • Sales and support: lead lists, personalised emails, SEO work.

Limits and caveats

  • Runtime-based billing means long tasks get expensive quickly; the free tier is only enough to try it out.
  • Automating sites that require a login may trigger anti-bot measures; check the target site's terms first.
  • The agent really runs commands and touches files, so when self-hosting, restrict container permissions and outbound network access.
  • Output still needs human review, especially financial figures and cited sources.

FAQ

Is Suna free?

There is a free tier with 10 minutes of runtime per month; Pro is $29 per month for 4 hours, Enterprise is $199 per month for 40 hours. Beyond that you upgrade or wait for the next cycle.

Can I self-host it?

Yes. Suna is open source, hosted on GitHub and designed for self-hosting; you can plug in your own OpenAI, Anthropic or other model keys.

How is it different from a normal chatbot?

A chatbot mostly returns advice and text. Suna actually drives a browser, reads and writes files and runs shell commands inside an isolated container, then hands back the finished result.

Frequently asked questions

Is Suna free?

There is a free tier with 10 minutes of runtime per month; Pro is $29 per month for 4 hours, Enterprise is $199 per month for 40 hours. Beyond that you upgrade or wait for the next cycle.

Can I self-host it?

Yes. Suna is open source, hosted on GitHub and designed for self-hosting; you can plug in your own OpenAI, Anthropic or other model keys.

How is it different from a normal chatbot?

A chatbot mostly returns advice and text. Suna actually drives a browser, reads and writes files and runs shell commands inside an isolated container, then hands back the finished result.