TL;DR: Kortix Suna is an open-source generalist AI agent you can self-host or use in the cloud to research the web, automate browsers, generate documents, analyze data, and run complex workflows—like a digital employee that follows plain-English instructions. It’s licensed under Apache-2.0, supports private, on-prem deployments, and offers tiered plans (including a free tier) based on usage time rather than feature lock-ins.
What is Kortix Suna?
Kortix Suna is a “generalist AI worker” built to act on your behalf—think of it as delegating real tasks (research, data wrangling, report building, web form filling, monitoring, etc.) instead of just getting an answer in chat. The official site positions it as an AI workforce you can spin up, customize, and deploy—fully open source and geared for real-world use.
Core idea: instead of a chatbot that explains how to do something, Suna actually does the thing—from browsing to file management to multi-step workflows.
Why Suna stands out
- Open-source, Apache-2.0 licensing (auditable code, forkable, modifiable). GitHub
- Self-hosting & data sovereignty: deploy entirely behind your firewall—no external dependencies required for your agents.
- End-to-end agent platform: browser automation, file/document ops, system commands, API integrations, and an agent builder—under one roof.
- 100+ integrations and support for premium model access via AI token credits. Suna
- Modern stack: TypeScript + Python; backend via FastAPI, frontend via Next.js, isolated Docker runtimes, Supabase for auth/storage.
Key capabilities (with real examples)
- Research & synthesis
Suna can browse multiple sources, extract information, and deliver a clean summary or competitive analysis—citing what it found. - Browser automation
Navigate sites, log in, fill forms, paginate, scrape, and submit—great for lead capture, vendor onboarding, or price monitoring. - Documents & content
Generate reports, proposals, contracts, presentations (PDF/Word/PowerPoint) and even visual assets to spec. - Data work
Upload spreadsheets/CSVs and ask for clean-ups, charts, forecasts, and KPI monitoring—then schedule repeat runs. - Workflow orchestration
Chain tasks: research → draft → design slides → email the deck; or monitor a site → scrape → append to a sheet → notify Slack. - System & DevOps tasks
Run safe CLI operations, manage configs, and automate build/deploy chores in sandboxed runtimes.
Pricing (usage-based by “AI Employee time”)
Kortix publishes clear, time-based tiers for its managed cloud:
- Free: $0 → 10 minutes / month (good for trials)
- Pro: $29 / month → 4 hours / month
- Enterprise: $199 / month → 40 hours / month
These are official figures shown on Kortix’s pricing page and may evolve—always check the live page.
The separate Plus/Pro/Ultra plans listed on Suna’s .so site reflect token-credit-style subscriptions for premium models and larger agent fleets. If your goal is raw “agent time,” the .lat page’s minutes/hours breakdown is the simplest way to plan monthly capacity.
Architecture (how Suna actually works)
- Backend API (FastAPI, Python) orchestrates agents, threads, tools and connects to LLMs (Anthropic, OpenAI, others via LiteLLM).
- Frontend (Next.js/React) gives you a dashboard for chat, configuration, workflow building, monitoring, and deployment.
- Agent Runtime (Docker) isolates each agent instance and provides the sandbox + browser automation + file system access + code interpreter.
- Database/Storage (Supabase) manages auth, file storage, conversation history, configs, analytics, and real-time monitoring.
Self-hosting: what you need to know
Quick start (from the official repo):
git clonethe repo → run the setup wizard (14 steps) →python start.pyto bring Suna online.
Typical dependencies:
- Supabase (auth + DB), Redis (caching/sessions), Daytona (secure execution), LLM keys (Anthropic/OpenAI via LiteLLM), optional Tavily/Firecrawl for search, QStash for background jobs. GitHub
A community walkthrough notes the setup is doable but non-trivial and, by default, Suna can call a frontier model like Claude Sonnet; your token spend will depend on the tasks you run. Treat this as a third-party anecdote rather than an official benchmark.
How Suna compares to typical “AI assistants”
Most chat assistants explain how to do something. Suna is explicitly built to do the thing—automate the browser, operate files, and chain steps end-to-end. That “hands-on” design is why Kortix calls it an AI worker/employee, not just a chatbot.
If you’ve tried “AI agent frameworks,” the difference here is production-ready packaging (dashboard + runtime + storage + integrations) rather than a pile of scripts.
Real-world use cases (you can copy)
- Sales & Growth: Prospecting → website visits → data capture → CRM update → personalized outreach sequence.
- Market/Competitor Research: Crawl public sources → summarize findings with citations → produce a report & slides.
- Ops Automation: Vendor onboarding forms → doc generation → email dispatch → status sheet update.
- Data & BI: Weekly ingest of CSVs → clean/merge → charts & KPIs → send PDF dashboard.
- Dev & IT: Script routine admin tasks in a sandbox; keep audit trails in Supabase.
Step-by-step: launch a pilot in one week
Day 1–2: Define 1–2 high-value workflows
Pick something repetitive but rule-based (e.g., compile competitor news → slide deck). Draft the steps in plain English.
Day 3: Cloud trial or lab server
Use the Free tier to validate the flow and token/time consumption. If self-hosting, complete the wizard (Supabase + Redis + Daytona).
Day 4–5: Build & chain tools
Configure browsing, file ops, and any required API integrations (e.g., Google Drive/Sheets/CRM).
Day 6: Guardrails & review
Limit domains, add check-points, require human approval before final publishing—Suna supports sandboxing and controlled execution.
Day 7: Measure & iterate
Log time saved, errors avoided, and deliverable quality. Increase “agent time” (hours) if ROI is positive.
Security, privacy, and governance
- On-prem only is supported for organizations needing complete data isolation; no external dependencies if you deploy the whole stack privately.
- Isolated runtimes with Docker help contain what an agent can access; combine with network egress rules and least-privilege API keys.
- Open codebase means you can audit, fork, and enforce your own policies and logs.
Performance notes (for builders)
Kortix also maintains Fast-Apply models for code-editing pipelines (separate repo). Reported dev-benchmarks show ~340 tok/s on a 1.5B model and ~150 tok/s on 7B in favorable hosting setups—useful if your agents do heavy “apply this diff” editing. Consider these directional, not SLA numbers.
Common pitfalls (and easy fixes)
- Confusing “Suna” with “Suno” (the AI music app). They’re unrelated. If you’re seeing “generate songs,” you’re on the wrong site. Suno is for music; Suna is an AI worker platform.
- Under-scoping permissions: Give agents only the keys and folders they need. Sandbox aggressively.
- Skipping observability: Turn on logs and review runs until you trust the automation. Supabase’s real-time layer helps.
- Estimating cost by tokens only: Cloud plans here meter by time (minutes/hours). Track time, not just tokens.
Who is Suna best for?
- Founders & operators who want concrete outputs (reports, decks, sheets) without adding headcount.
- Ops/RevOps teams automating repetitive web + doc + spreadsheet tasks.
- Security-conscious orgs that prefer open-source and on-prem control.
Quick start links & proof points
- Suna website (.so)—capabilities, token-credit plans, and on-prem messaging.
- Suna website (.lat)—usage-time pricing (minutes/hours).
- GitHub repo—README, architecture, Quick Start, releases (v1 on Apr 22, 2025), languages (TS/Python), and star/fork counts.
- Dev community guide—independent setup/deployment write-up with notes on model defaults and deps.
- Fast-Apply repo—Kortix’s code-editing pipeline models & throughput notes.
FAQs (high-intent)
Is Suna really free?
Yes, there’s a Free cloud tier with 10 minutes/month—enough to test a couple of workflows. Self-hosting is free under Apache-2.0 (infra/model costs are yours).
Can I run Suna fully on-prem?
Yes. Kortix provides a self-hosting wizard and documents running everything behind your firewall, including Supabase, Redis, and sandboxed runtimes.
Which LLMs power Suna?
You can connect multiple providers (OpenAI, Anthropic, etc.) via LiteLLM. Choice affects cost, latency, and quality—start with what your security team approves.
How is Suna different from “agent frameworks”?
It ships as a complete platform—UI, runtime, storage, integrations—so non-ML teams can operate real agents without gluing many tools together.
How does pricing compare to token-based AI apps?
Kortix’s cloud pricing is time-based (minutes/hours of “AI employee” work). If you self-host, your cost is infra + model API usage.
Verdict: Should you “hire” Kortix Suna?
If you want an auditable, self-hostable, production-ready agent that does actual work (not just chat), Suna is easy to recommend. Start small on the Free tier, estimate your time needs, then scale to Pro/Enterprise or go on-prem if you need total control. For teams with privacy requirements—or anyone ready to move from “answers” to outputs—Suna is one of the most compelling open-source options today.
Note: Don’t mix up Suna (Kortix’s AI worker platform) with Suno (AI music generation app). Different products, different goals.
If you’d like, I can tailor a pilot workflow for your exact use case (e.g., weekly market scan → slide deck → stakeholder email) and include prompts, guardrails, and a rollout checklist.
Subscribe to our channels at alt4.in or at Knowlab
