8 tools compared on real pricing, genuine strengths, and the limitations that only show up after you buy.
We track 8 Agent Development tools, 8 of which have a free or freemium tier you can evaluate without a sales call. Our picks are LangGraph, CrewAI and Haystack. Each entry below lists real pricing, what it is genuinely good at, and the limitations that only show up after you buy.
| Tool | Pricing model | What it costs |
|---|---|---|
| LangGraph | freemium | LangGraph itself is open source under the MIT licence and free to use. Managed hosting is LangSmith Deployment: the LangSmith Plus plan is $39 per seat per month (billed monthly), includes 10,000 base traces a month (then 0.005 LSU per base trace) and one free Small serverless deployment; additional Serverless and Dedicated deployments are metered in LSUs ($1.00 each) at 0.0675 LSU per runtime vCPU-hour, 0.0090 LSU per runtime GiB-hour, 0.177 LSU per database vCPU-hour and 0.025 LSU per database GiB-hour, so a Small serverless deployment (0.8 vCPU, 2 GiB) costs 0.072 LSU per running hour. Self-hosted and hybrid deployment are Enterprise options at custom pricing. As read on the vendor's pricing page on 10 November 2026. |
| CrewAI | freemium | The CrewAI framework is open source under the MIT licence and free. The CrewAI platform has a free Basic plan with 50 workflow executions per month and 2 automations; Enterprise is custom-priced with "flexible overage", SSO, RBAC, and deployment on CrewAI cloud, your own VPC or your own infrastructure. No paid self-serve tier or overage rate is published. As read on the vendor's pricing page on 10 November 2026. |
| Haystack | freemium | Open-source framework is free. The deepset AI Platform is a paid, enterprise-oriented managed offering. |
| LlamaIndex | freemium | The LlamaIndex framework is MIT-licensed and free; the company says its primary focus has shifted to LlamaParse, its paid document-processing platform, and the TypeScript port LlamaIndex.TS is deprecated. As read on the project's GitHub README on 10 November 2026. |
| Microsoft Agent Framework | free | The framework is MIT-licensed and free. Hosting is optional and separate: on Foundry Agent Service there is "no additional charge for creating or running Foundry-native agents using prompts and workflows", while hosted agents bill container compute at $0.0994 per vCPU-hour and $0.0118 per GiB-hour (West US 3), plus model tokens and tool fees such as Web Search at $14 per 1,000 transactions. As read on the vendor's pricing page on 11 November 2026. |
| Microsoft AutoGen | free | AutoGen's code is MIT-licensed (documentation CC-BY-4.0) and free, but its README declares the project archived and no longer maintained: Microsoft provides no bug fixes, security fixes, new features or support, advises against production use, and directs users to Microsoft Agent Framework. As read on the project's GitHub README on 10 November 2026. |
| OpenAI AgentKit | freemium | The Agents SDK (Python and JS/TS) is MIT-licensed and free; you pay OpenAI API model rates plus built-in tool fees: web search $10 per 1,000 calls plus search content tokens, file search $2.50 per 1,000 calls plus $0.10 per GB-day storage (1 GB free), code containers from $0.03 per 20-minute session (1 GB; eligible sessions billed by the minute with a 5-minute minimum), and ChatKit upload storage $0.10 per GB-day (1 GB free). OpenAI's managed Agents API (durable sessions on OpenAI infrastructure) has no separate platform fee: model, tool and container rates apply. Agent Builder is deprecated and scheduled to shut down on 30 November 2026; ChatKit remains available. As read on the vendor's pricing page on 10 November 2026. |
| Semantic Kernel | free | Semantic Kernel is MIT-licensed and free, and Microsoft now describes Microsoft Agent Framework as its "enterprise-ready successor" (version 1.0 released 3 April 2026, also MIT). Its successor, Microsoft Agent Framework, can be hosted on Foundry Agent Service: no added charge for Foundry-native prompt and workflow agents; hosted agents bill $0.0994 per vCPU-hour and $0.0118 per GiB-hour in East US, plus model tokens and tool fees (for example Web Search at $14 per 1,000 transactions). As read on the vendor's pricing page on 10 November 2026. |
A low-level orchestration framework for building stateful, multi-step agent workflows with explicit control over state, branching, and human-in-the-loop steps.
A framework for orchestrating role-based multi-agent teams, where specialized agents collaborate on a task under a defined process.
An open-source framework from deepset for building production LLM applications — RAG, search, and agents — built around composable pipelines.
A data framework for connecting LLMs to private and enterprise data — ingestion, indexing, retrieval, and agent workflows over your own content.
Microsoft's open-source agent framework for Python and .NET, the 1.0 successor to both Semantic Kernel and AutoGen, with graph-based workflows and hosting on Foundry Agent Service.
Microsoft Research's open-source multi-agent framework, now archived: Microsoft no longer fixes, secures or supports it and points users to Microsoft Agent Framework.
OpenAI's first-party agent toolkit: the open-source Agents SDK, the embeddable ChatKit UI and evals, billed at plain API rates. Agent Builder shuts down on 30 November 2026.
Microsoft's open-source SDK for integrating LLMs into C#, Python and Java applications, now succeeded by Microsoft Agent Framework (1.0, April 2026).
Practical playbooks, prompts, and tool stacks — for the SMB operators and enterprise teams actually shipping AI.
Free implementation guides for SMB operators and enterprise teams — workflows, prompts, governance, and tool stacks built to ship.