Contents
- Verdict
- What is Agno?
- Best for, and not for
- Strengths and limits
- Pricing
- What people actually build with Agno
- What Agno actually costs at small-business volume
- The gotcha: free to serve is not free to run
- What breaks in production, and how you find out
- What it connects to that you already run
- Agno alternatives: what to pick instead, and when
- Migration and lock-in: can you get your work out?
- FAQ
Agno has some of the least metered pricing in this directory. The framework and the AgentOS runtime are free and open source, the pricing page says serving costs nothing on any plan, and the only paid product is the control plane: $150 a month once you point it at a live deployment. A thousand runs a month and a million cost the same in Agno fees, which is to say nothing beyond that flat charge. What you pay instead is your own hosting, your own PostgreSQL and your model tokens, plus the time to operate them. Worth it for a Python team that wants its data in its own cloud. The wrong fit for a business that wanted someone else to run the thing.
What is Agno?
Agno is an open-source Python framework for building agents, teams of agents and workflows, together with AgentOS, a runtime that serves those agents as an API from your own infrastructure. Agno's documentation describes AgentOS as handling authentication, role-based access, human approvals, scheduling and durable execution, with PostgreSQL as the store for conversations and memory. The company's paid product is a hosted control plane, a web interface for chat, sessions, traces, memory and evaluations, that connects to an AgentOS you run. It is free against a local AgentOS and paid when pointed at a live deployment.
Best for
Python teams that want to host agents in their own cloud and pay a flat fee, not a usage meter, for the management interface.
Not for
Teams without Python developers, or anyone who wants the vendor to host and run the agents.
Strengths
- The framework and the AgentOS runtime are open source under Apache 2.0
- The pricing page states that serving costs nothing on any plan, so there is no per-run or per-token charge from Agno
- Agents run in your own cloud, so conversations and memory sit in your own PostgreSQL database
- The full control plane is free with unlimited use against a local AgentOS
- Paid pricing is flat and fully published: $150 a month, $30 a seat, $95 a connection, $300 for SAML SSO
- A very large and active repository, pushed on the day we checked
Limitations
- You host, scale, secure and back up the runtime and its database yourself
- The control plane is only free against a local AgentOS; a live deployment needs the $150 a month Pro plan
- Each extra live connection is $95 a month, and the pricing page does not define what counts as one connection
- SAML single sign-on is a $300 a month add-on, twice the base plan
- The pricing page does not describe what data passes through Agno's hosted control plane
- Python only, and no visual builder for non-developers
Agno pricing
Agno charges nothing to build or serve agents: the SDK and AgentOS are open source and run in your own cloud. What it sells is the control plane. Free uses it against a local AgentOS with unlimited use. Pro is $150 a month and connects it to one live AgentOS with three team seats, role-based access control and email support. Extra seats are $30 a month, extra connections $95 a month and SAML SSO $300 a month. Enterprise is custom.
| Plan | Price | What you get |
|---|---|---|
| Free | $0/mo | Build and run your agent platform in your own cloud at any scale. Full control plane against a local AgentOS with unlimited chat, sessions, traces, studio, memory, knowledge, metrics, evaluations, approvals and scheduler |
| Pro | $150/mo | Connects the control plane to a live deployment: one live AgentOS connection, three team seats, role-based access control, email support |
| Pro add-ons | $30/seat | Extra seats $30 a month each, extra live connections $95 a month each, SAML SSO $300 a month |
| Enterprise | Custom | End-user identity, roles and audit logs, custom RBAC with per-resource scoping, security reviews, monthly office hours and dedicated build help |
| Runtime and SDK | $0 | Open source under Apache 2.0. The pricing page states that serving costs nothing on any plan. You pay your own hosting, database and model providers |
Pricing reflects public plans as of November 17, 2026 and can change. Check Agno for the latest.
What people actually build with Agno
Agno is aimed at teams that want to run agents as a service of their own rather than rent one. The framework builds the agents; AgentOS serves them from your infrastructure; the control plane lets people watch and manage them. The builds that suit it need that whole shape.
The builds that fit
An internal agent platform for a company. Several agents, each with its own tools and knowledge, served from one runtime in the company's own cloud. The documentation lists deployment targets including Docker, Kubernetes, AWS, GCP, Azure, Render, Railway, Fly and Modal.
Inbox triage at our reference volume. An agent that classifies and extracts from 1,000 messages a month, on a schedule, with results stored in your own database. Because serving is free, the Agno bill for this is zero if you manage it locally and $150 if you want the hosted control plane on the live deployment.
Agents with memory about users. The documentation describes memory that remembers users across sessions and knowledge retrieval from your own data, with PostgreSQL holding conversations and platform activity. For a support or account assistant, that data staying in your own database is the point.
Workflows with human approval. AgentOS is described as handling human approvals, scheduling and durable execution, and workflows support routing, conditions, loops and parallel steps. That suits processes where an agent drafts and a person signs off.
Agents exposed in several places at once. The pricing page says agents can be served over REST, to Claude and ChatGPT through MCP, and in Slack and Discord, from the same runtime.
What it is genuinely bad at
Being hands-off. Somebody has to deploy the runtime, run the database, apply upgrades, watch for failures and keep it secure. For a small business with no engineer, that is the entire cost and it is not on the pricing page. It is also Python only, with no visual builder, so a non-technical owner cannot create or change an agent.
Bottom line: Agno fits a Python team that wants to own an agent service inside its own cloud, and does not fit a business looking for someone else to operate one.
What Agno actually costs at small-business volume
Agno's pricing page, read on 17 November 2026, has three plans and one sentence that matters more than any of them: serving costs nothing on any plan. The SDK and AgentOS are open source. There is no per-run, per-agent, per-message or per-token charge from Agno listed anywhere on the page.
So for our reference workload of 1,000 triage runs a month, the Agno fee is the same as for 1,000,000: nothing on Free, and $150 a month on Pro. Volume does not enter into it. That makes Agno unlike almost every other entry in this directory, where the bill is a function of usage.
Free covers building and running the platform in your own cloud at any scale, with the full control plane working against a local AgentOS: chat, sessions, traces, studio, memory, knowledge, metrics, evaluations, approvals and the scheduler, all unlimited. Pro, at $150 a month, is what connects the control plane to a live deployment, and includes one live AgentOS connection, three team seats, role-based access control and email support.
The add-ons are flat and published. Extra seats are $30 a month. Extra connections are $95 a month. SAML single sign-on is $300 a month.
What Agno does not bill, you pay elsewhere. A small always-on server and a managed PostgreSQL instance at your cloud provider, and the model tokens for each run at your provider's rates. Those are your contracts, at whatever you have negotiated.
| Configuration | Base | Seats | Connections | SSO | Monthly total |
|---|---|---|---|---|---|
| Local AgentOS only, any team size | $0 | - | - | - | $0 |
| 3 people, 1 live deployment | $150 | Included | Included | - | $150 |
| 5 people, 1 live deployment | $150 | 2 x $30 = $60 | Included | - | $210 |
| 5 people, 2 live deployments | $150 | 2 x $30 = $60 | 1 x $95 = $95 | - | $305 |
| 5 people, 2 live deployments, SSO | $150 | $60 | $95 | $300 | $605 |
| 10 people, 3 live deployments | $150 | 7 x $30 = $210 | 2 x $95 = $190 | - | $550 |
| 10 people, 3 live deployments, SSO | $150 | $210 | $190 | $300 | $850 |
Three realistic configurations, with the arithmetic
A three-person team with one production deployment pays the base plan: $150 a month, $1,800 a year.
A five-person team with separate staging and production deployments, each connected, pays $150 plus two extra seats at $30 ($60) plus one extra connection at $95. That is $305 a month. Add SAML SSO and it is $605 a month, $7,260 a year.
A ten-person team with three connected deployments pays $150 plus seven seats at $30 ($210) plus two connections at $95 ($190), which is $550 a month, or $850 with SSO.
The pattern is that seats are cheap and connections and SSO are not. The pricing page does not define a connection beyond the phrase one live AgentOS connection, so whether a staging environment needs its own is a question to settle with Agno before budgeting. If staging can be managed locally on the free tier, the five-person example drops to $210.
Bottom line: Agno charges nothing per run, so 1,000 runs and a million cost the same: $0 managed locally, or a flat $150 a month to connect the control plane to one live deployment with three seats.
The gotcha: free to serve is not free to run
Agno's headline is accurate and incomplete in the way self-hosted software always is. Serving costs nothing in Agno fees. It costs you a server, a database and a person.
The server and the database are small and easy to price at your own provider. The person is the real line. Someone has to deploy AgentOS, keep it patched, back up the PostgreSQL database that holds every conversation and memory, rotate credentials, watch for failures and handle upgrades to a framework that is changing quickly. For a team that already runs services, that is a few hours a month added to an existing routine. For a business with no engineer, it is the whole project, and no plan on the pricing page covers it.
The second thing to read carefully is where the free tier ends. The control plane is free against a local AgentOS. The moment you want to use that same interface on a deployment your customers or staff actually use, you are on Pro at $150 a month. That is a fair place to draw the line, and it means the free tier is a development tool, not a way to run production with a hosted dashboard for nothing. You can of course run production without the hosted control plane at all, using your own logging, and pay Agno nothing.
Third, the add-ons are large relative to the base. A second connection adds 63 percent to the base price. SAML SSO at $300 a month is twice the plan it attaches to. A company whose security policy requires single sign-on is really looking at $450 a month as its starting price, not $150.
Fourth, an open question rather than a charge. The pricing page and the documentation introduction say you keep your data in your control by running the platform in your own cloud, and that you connect your platform to Agno's hosted interface to use it. Neither page we read describes what the hosted control plane sees when it is connected to a live deployment. For most teams that is a question to ask; for a regulated one it is a question to have answered in writing before connecting anything.
Bottom line: Agno bills nothing for serving, so the real costs are your hosting, your database and the engineer who operates them, and a company that needs SSO starts at $450 a month, not $150.
What breaks in production, and how you find out
With Agno, production reliability is yours by design, so the failure modes are those of any service you host. The runtime goes down and nobody is paged. The database fills up. A deploy drops in-flight runs. An upgrade changes behaviour. None of these is Agno's fault and all of them are your job.
The database deserves particular attention. The documentation says storage is PostgreSQL, used for conversations, memory and platform activity. That makes it the single most valuable component in the system: it holds what your agents know about your users. Back it up, test the restore, and decide how long conversations are kept, because a memory store that grows forever becomes both a cost and a liability.
On project health, the GitHub API on 17 November 2026 showed 43,214 stars, 1,880 open issues and a push that same day. That is a large and very active project, and the open issue count reflects its size and pace. The practical consequence is the usual one for a fast-moving framework: pin versions, read release notes, and keep a set of saved conversations to re-run after every upgrade so you notice a behaviour change before a customer does.
AgentOS is described as providing durable execution, resume and replay, human approvals and scheduling. Those are the features that make an agent safe to leave running, and they only help if you use them. Put an approval step in front of anything that sends a message, moves money or changes a customer record. Schedule work through the runtime rather than an external cron so that failures are visible in one place.
The control plane's traces, metrics and evaluations are the monitoring. Against a local AgentOS they are free, so there is no reason not to run evaluations during development. In production, on Pro, they are how you find out that an agent has started answering differently.
Security and compliance, as published
Role-based access control is included on Pro and SAML SSO is a $300 a month add-on. End-user identity, audit logs, per-resource RBAC scoping and security reviews are on Enterprise. The pricing page does not mention SOC 2, HIPAA or data residency, and we are not inferring them. Because the runtime and data live in your own cloud, most of your compliance position is determined by your own infrastructure and your model provider.
Bottom line: Agno's runtime is yours to operate, so the things that break are a host nobody monitors and a PostgreSQL database nobody backed up.
What it connects to that you already run
Agno's integrations fall into three groups. The first is where agents are exposed. The pricing page says agents can be served over REST, to Claude and ChatGPT through MCP, and in Slack and Discord. REST makes an agent callable from any application or automation platform you already run. MCP makes it usable as a tool from inside the assistants your staff already have open.
The second is what agents can reach. As with every code framework, tools are functions you write or import, so your CRM, database and internal services are reached in Python. There is no drag-and-drop connector catalogue. The documentation also describes teams of agents that use tools, delegate work and collaborate, and knowledge retrieval over your own data.
The third is infrastructure. PostgreSQL is the store, which most businesses with any engineering already run or can get as a managed service from their cloud. Deployment targets listed in the documentation cover the mainstream options, from a single Docker container to Kubernetes, so AgentOS goes wherever the rest of your services are.
The documentation also names the Agno, Claude Agent and Codex SDKs for building agents, which suggests the runtime is not limited to agents written with Agno's own framework.
For a small business the practical pattern is to expose each agent over REST and let an automation platform such as n8n or Make.com call it as one step in a larger flow. The agent does the judgement; the automation platform does the plumbing to the systems of record, where a non-developer can see it. That keeps the amount of Python you maintain small.
Bottom line: Agno agents are reached over REST, MCP, Slack and Discord and reach your systems through Python tools, with PostgreSQL and your own cloud underneath.
Agno alternatives: what to pick instead, and when
The question that sorts the alternatives is who runs the runtime. Agno says you do. If that is the wrong answer for your team, look at a managed runtime. Amazon Bedrock AgentCore runs agents built with any framework and bills compute by the second, with identity, memory and a tool gateway as separate metered services. You trade a flat $150 for a usage bill and no servers.
If you want a framework but in TypeScript, Mastra is the counterpart, with its own optional hosted platform from $0 to $250 a month. If you want a Python framework without a runtime or control plane attached, LangGraph gives explicit control over state and branching, CrewAI is organised around roles and teams, and Pydantic AI emphasises typed, validated outputs. All three are libraries you host yourself.
If your need is reliable execution of long runs rather than a framework, Trigger.dev provides that with per-second billing, though in TypeScript. And if nobody on the team writes code, choose a visual builder such as n8n instead of any of these.
| Tool | Who runs it | What you pay the vendor | Who it suits | Does better than Agno |
|---|---|---|---|---|
| Agno | You, in your own cloud | $0 to serve; $150 a month for the control plane on a live deployment | Python teams wanting data in their own cloud | Flat pricing with no usage meter |
| Amazon Bedrock AgentCore | AWS | $0.1276 a vCPU-hour and $0.0169 a GB-hour on Runtime v2, plus per-service fees | Teams already on AWS | No servers to operate |
| Mastra | You, or Mastra's platform | $0 for the framework; platform $0, $25 or $250 a month | TypeScript product teams | A hosted option from the same vendor |
| LangGraph | You | Free library (August review) | Teams wanting explicit graph control | Fine control of state and branching |
| CrewAI | You | Free library (August review) | Role-based multi-agent designs | A simple mental model for agent teams |
| Pydantic AI | You | Free library (August review) | Teams wanting typed, validated outputs | Type safety and a small surface |
| n8n | n8n cloud or you | Per-execution plans (August review) | Teams without developers | A visual builder with connectors |
Bottom line: Choose Agno if you want to run the agent service yourself for a flat fee, and a managed runtime if you would rather pay by usage than operate servers.
Migration and lock-in: can you get your work out?
Agno's lock-in position is strong for a structural reason: almost everything is already yours. The agents are Python in your repository. The runtime is open-source software on your own servers. The conversations, memory and activity are rows in your own PostgreSQL database. If Agno the company changed its pricing or disappeared, the system you are running would keep running.
The licence supports that. The GitHub API reports the repository as Apache 2.0, a permissive licence that allows commercial use, modification and distribution. That is a simpler position than frameworks whose repositories mix licences, and it means the runtime cannot be repriced out from under you.
What you would lose by dropping the paid plan is the hosted control plane on your live deployment: the web interface for sessions, traces, evaluations and approvals, and the seats and access control attached to it. Pro is listed monthly, so that is a cancellation rather than a negotiation. You would replace it with your own logging and dashboards, or with the same control plane used locally.
What you would rewrite to leave the framework is what you rewrite to leave any framework: agent, team and workflow definitions written against Agno's interfaces. The durable assets carry over. The instructions, the tool functions, the evaluation cases and above all the data, which is already in a standard database you control and can query with ordinary SQL.
The one piece of advice specific to Agno is to treat the database as the system of record from the first day. Know its schema, back it up, and do not let the only readable view of your agents' memory be a hosted interface.
Bottom line: With Agno the code, the Apache 2.0 runtime and the PostgreSQL data are all on your side already, so the only thing a cancellation removes is the hosted control plane.
Agno FAQ
Is Agno free?
The framework and the AgentOS runtime are open source and free to run in your own cloud at any scale, and the control plane is free against a local AgentOS. Connecting the control plane to a live deployment requires Pro at $150 a month, as published on 17 November 2026.
How much does Agno Pro cost?
$150 a month, including one live AgentOS connection, three team seats, role-based access control and email support. Extra seats are $30 a month each, extra connections $95 a month each and SAML SSO $300 a month.
Does Agno charge per run or per token?
No such charge is listed. The pricing page states that serving costs nothing on any plan. Model tokens are billed by your model provider, and hosting by your cloud provider.
What is AgentOS?
The runtime that serves your agents as a service. Agno's documentation says it handles authentication, role-based access, resume and replay, human approvals, scheduling and durable execution, and that it runs in your own cloud with PostgreSQL as its store.
What does a five-person team pay?
With one live deployment: $150 plus two extra seats at $30, so $210 a month. With a second connected deployment add $95, for $305. With SAML SSO as well add $300, for $605 a month.
What licence is Agno under?
The GitHub API reports the agno-agi/agno repository as Apache 2.0, a permissive open-source licence that allows commercial use and modification.
Where does my data live?
In your own infrastructure. The documentation says you run the platform in your own cloud and keep your data in your control, with PostgreSQL storing conversations, memory and platform activity. The pages we read do not describe what the hosted control plane accesses when connected, so ask Agno if that matters to you.
Does Agno host my agents for me?
No plan on the pricing page includes hosting. You deploy AgentOS yourself, to targets the documentation lists such as Docker, Kubernetes, AWS, GCP, Azure, Render, Railway, Fly or Modal.
Is single sign-on included?
No. SAML SSO is a $300 a month add-on to Pro. Role-based access control is included in Pro, and finer-grained access controls and audit logs are on Enterprise.
Looking at alternatives? Agno hands a Python team a capable runtime and expects them to host and operate it. Autonoly is the opposite arrangement: a managed service where the automation is built and run for you around the tools you already use. If you have Python developers and want agents inside your own cloud, Agno is a strong choice at a predictable price. If you have no developers, a free runtime is not free. See the Autonoly review.
