</>macrostackBrowse all
Migration guide · AI agent frameworks

The 4 best OpenAI AgentKit alternatives

AgentKit was OpenAI's managed way to build agents — Agent Builder for wiring the workflow visually, ChatKit for the chat surface, and Evals for measuring whether the thing worked. The pitch was that you would not need a framework at all: describe the agent, connect the tools, ship. It arrived as the safe, official option in a category otherwise full of fast-moving open-source projects.

92
Bottom line

Migrate before November 2026, and this time pick something nobody can switch off. LangGraph is the closest match for the workflows AgentKit was built for and it is MIT-licensed code in your own repository — if it is ever abandoned, you still have it.

Jump to the full comparison →

The cost

There was no separate subscription — you paid ordinary API rates for what the agent consumed, roughly $1.75 per million input tokens and $14 per million output tokens on the flagship model, plus tool charges such as $0.03 per Code Interpreter session and $0.10 per gigabyte per day for file search. Enterprise agreements sat between roughly $2,000 and $20,000 a month. Figures checked 2026-07-30. The pricing was never the problem.

Why people consider an alternative

Because it is being switched off. On 3 June 2026 OpenAI announced it was winding down Agent Builder and Evals, with the products leaving the platform on 30 November 2026. Nothing was wrong with the code and nobody was priced out — the vendor simply changed direction, and every team that built on it now has a migration with a date attached. That is the whole argument this site exists to make, demonstrated by the largest company in the industry inside a single year: the risk in a managed platform is not usually the bill, it is that the roadmap is not yours. It is worth being fair here — OpenAI gave nearly six months of notice, which is more than many vendors manage. The point is not that they behaved badly. The point is that the decision was never yours to make.

When OpenAI AgentKit is still the right call

Through November 2026, if what you have already works and rewriting it now costs more than running it out — there is no benefit in a panic migration for a system that will be replaced anyway. Stay too if you are prototyping this week and expect the result to be thrown away: managed tooling is genuinely faster for that, and a throwaway prototype has no migration cost by definition. And for the rest of OpenAI's API — models, ChatKit, the Responses API — nothing here is a reason to leave. Only the agent-building layer is closing.

The verdict — at a glance

AlternativeLicenseSelf-hostPricingSovereignty
LangGraph★MITYesFree and open source. LangGraph Platform is an optional paid hosted tier.92
OpenClawMITYesFree and open source. You pay only for whatever model you point it at.94
CrewAIMITYesFree and open source. CrewAI Enterprise is a separate commercial platform.90
DifyModified Apache-2.0 (source-available, not OSI open source)YesFree to self-host within the licence terms. Dify Cloud and a commercial licence are paid.72
92
Macrostack's top pick

LangGraph

Agents as an explicit state graph — the closest thing to what you are leaving.

Which one fits your situation?

There is no single best answer — it depends on what you are actually trying to do.

Just tell me the best OpenAI AgentKit alternative

LangGraph92

Our top pick — the highest Sovereignty Score at 92, and the one we would choose ourselves.

LangGraph vs OpenClaw →

I have no budget — is there a genuinely free one?

OpenClaw94

Open source under MIT. Free and open source. You pay only for whatever model you point it at.

OpenClaw vs LangGraph →

I want to self-host it and own the data

CrewAI90

Self-hostable, so your data stays on your own server. Free and open source. CrewAI Enterprise is a separate commercial platform.

CrewAI vs LangGraph →

It is just me — I do not want to run a server

Dify72

Offers a hosted option, so you can start without infrastructure. Free to self-host within the licence terms. Dify Cloud and a commercial licence are paid.

Dify vs LangGraph →

Every alternative, compared

#1★ TOP PICK

LangGraph

Agents as an explicit state graph — the closest thing to what you are leaving.

92
OPEN SOURCEMITSELF-HOSTLOCAL-FIRST

LangGraph models an agent as a graph of nodes with explicit state, checkpointing and human-in-the-loop interrupts, which is very nearly the mental model Agent Builder gave you — with the difference that the graph is code in your repository. It is the most battle-tested option for production agents that must survive restarts, resume mid-run and be inspected when they go wrong. MIT-licensed, roughly 38.5k GitHub stars and pushed to daily as of 2026-07-30. LangGraph Platform is a paid hosted deployment tier, but the framework itself is complete without it — the distinction is worth checking before you build on anything you assume is free.

Strengths

  • +Explicit state graph maps closely onto what Agent Builder did
  • +Checkpointing and resumable runs — genuinely production-grade
  • +Human-in-the-loop interrupts are a first-class feature, not a workaround
  • +MIT and in your repository: nobody can announce a sunset date for it

Trade-offs

  • −Steepest learning curve here — you are writing the graph, not drawing it
  • −Carries LangChain ecosystem conventions whether or not you want them
  • −Verbose for simple agents that only need a loop and two tools
  • −The hosted platform is where the operational conveniences live
Free and open source. LangGraph Platform is an optional paid hosted tier.
WebsiteSourceDocs
#2

OpenClaw

The runaway story of 2026 — 384k stars and MIT-licensed.

94
OPEN SOURCEMITSELF-HOSTLOCAL-FIRST

OpenClaw is a personal AI assistant that runs on your own machine, connects to whatever model you point it at, and executes real tasks across your OS. It went from roughly 9k GitHub stars to over 384,000 during 2026, overtaking React to become one of the most-starred repositories in history — a rate of adoption that no framework in this category has matched. It is MIT-licensed under the OpenClaw Foundation rather than a company, which matters for the specific risk that brought you to this page. It is a different shape from LangGraph and CrewAI: an assistant you run rather than a library you build with, so it fits personal and desktop automation far better than a multi-tenant production service.

Strengths

  • +MIT under a foundation, not a company — no single owner to change direction
  • +Runs locally and works with any model, including local ones
  • +The fastest-growing project in the category by an enormous margin
  • +Genuinely useful out of the box rather than a library you must assemble

Trade-offs

  • −An assistant, not a framework — the wrong shape for a hosted product
  • −Moving extremely fast; interfaces change under you
  • −Executing real actions on your machine deserves real caution about permissions
  • −Governance is young; the foundation is newer than the star count suggests
Free and open source. You pay only for whatever model you point it at.
WebsiteSource
#3

CrewAI

Agents as a team with roles — the fastest thing here to get working.

90
OPEN SOURCEMITSELF-HOSTLOCAL-FIRST

CrewAI frames the problem as staffing rather than graph theory: you define agents with a role, a goal and a backstory, hand the crew a task, and it coordinates them. That abstraction is the lowest barrier to entry in the category and it is why CrewAI has the most stars of the pure frameworks here — around 56.4k on 2026-07-30 — with roughly 5.2 million monthly downloads. It is independent of LangChain, which some teams specifically want. The trade is control: when the crew does something strange, there is more framework between you and the reason than there is in LangGraph.

Strengths

  • +Fastest path from idea to a working multi-agent system
  • +Role-and-goal abstraction is genuinely intuitive to reason about
  • +Standalone — no LangChain dependency to inherit
  • +Large, active community and plenty of worked examples

Trade-offs

  • −Less control over execution than an explicit graph
  • −Harder to debug when a crew misbehaves — the abstraction hides the path
  • −Weaker state persistence story for long-running work
  • −Company-controlled, with an enterprise tier alongside it
Free and open source. CrewAI Enterprise is a separate commercial platform.
WebsiteSourceDocs
#4

Dify

The visual builder AgentKit was — but read the licence before you commit.

72
SOURCE-AVAILABLEModified Apache-2.0 (source-available, not OSI open source)SELF-HOSTLOCAL-FIRST

Dify is the closest replacement for Agent Builder specifically: a visual canvas for wiring prompts, tools, retrieval and branching, with a hosted-app layer on top, so a team that liked drawing the workflow does not have to start writing graphs in Python. It is by far the most-starred project in this comparison at around 150.8k. One important correction to how it is usually described. Dify is very widely listed as open source, and its own licence file is headed "Open Source License", but the text is a modified Apache-2.0 with two additional conditions: you may not use the source to operate a multi-tenant service without written authorisation, and you may not remove or alter Dify's logo and copyright notices in the frontend. Those restrictions mean it does not meet the Open Source Definition. It is source-available. For most internal use that changes nothing; if you intend to build a product for your own customers on top of it, it changes everything, and it is exactly the clause people discover late.

Strengths

  • +The nearest thing to Agent Builder's visual canvas — smallest conceptual jump
  • +Self-hostable, with retrieval, tools and app hosting in one product
  • +Enormous community and the most stars of anything in this comparison
  • +Non-engineers can read and change the workflow, which is rare here

Trade-offs

  • −Not open source despite near-universal description as such — multi-tenant use needs a commercial licence
  • −Branding and logo may not be removed from the frontend
  • −Visual workflows are harder to review, diff and version than code
  • −Single-vendor control, and the licence explicitly reserves the right to tighten
Free to self-host within the licence terms. Dify Cloud and a commercial licence are paid.
WebsiteSourceDocs

Questions people ask

▸ Is OpenAI AgentKit being discontinued?

Yes. OpenAI announced on 3 June 2026 that Agent Builder and Evals are being wound down, and they leave the platform on 30 November 2026. ChatKit and the underlying API models are not affected. If you built on Agent Builder, you have a migration with a fixed date.

▸ What should I migrate from AgentKit to?

LangGraph if you want the closest match to Agent Builder's stateful workflows with production-grade checkpointing — it is MIT and lives in your repository. CrewAI if you want the fastest route to something working and think in terms of roles rather than graphs. Dify if what you actually valued was the visual canvas, with the caveat that its licence is not truly open source.

▸ Is Dify open source?

Not by the Open Source Definition, despite being described that way almost everywhere. Its licence is a modified Apache-2.0 that forbids operating a multi-tenant service without written authorisation and forbids removing Dify's branding from the frontend. It is source-available. Fine for internal use; a real constraint if you plan to build a product on it for your own customers.

▸ Is AutoGen still a good choice?

It is no longer the obvious one. Microsoft merged AutoGen and Semantic Kernel into Microsoft Agent Framework 1.0 in April 2026 and moved AutoGen itself to maintenance mode — its repository was last pushed on 2026-04-15. The conversation patterns are still the most diverse in the category, but building new work on a project in maintenance repeats the mistake that brought you to this page.

Compare them head-to-head

LangGraph vs OpenClaw→LangGraph vs CrewAI→LangGraph vs Dify→OpenClaw vs CrewAI→OpenClaw vs Dify→CrewAI vs Dify→

Related comparisons

Alternatives to OpenAI API (ChatGPT)8 optionsAlternatives to Notion4 optionsAlternatives to Datadog4 optionsAlternatives to 1Password4 optionsAlternatives to Google Analytics4 optionsAlternatives to Slack3 options

Entry last verified 2026-07-30. Licenses and pricing change — spotted something out of date? That's a correction we want.

The Macrostack brief

New swaps, worth your inbox.

A short, occasional email when we add a high-intent alternative or ship a new head-to-head. No spam, no selling your address — unsubscribe in one click.