Key facts
| CrewAI | Open-source Python framework with crews and flows |
| CrewAI agents | Role, goal and backstory define each agent's behaviour |
| CrewAI models | Bring your own LLM endpoint and keys |
| Plugsky | OpenAI-compatible API with live function calling and 30+ models |
| Running together | Point the CrewAI model client at Plugsky's base URL |
| Platform controls | Scoped keys, RBAC, SSO/SCIM and audit logs are live |
| Deployment | Cloud, VPC, on-prem and air-gapped with region choice |
| Honest limit | CrewAI's framework features are not replaced by an API |
TL;DR
- CrewAI is orchestration code you host; Plugsky is the model runtime.
- Most teams keep CrewAI and swap the model layer only.
- One key and 30+ models make routing between cheap and frontier tiers simple.
- You still own crew design, state, tooling and evaluation.
- OpenAI-compatible interfaces keep your exit cost low.
How it works, step by step
- Inventory what CrewAI does for you: crews, flows, tools, memory and callbacks.
- Check which LLM client the crews use and whether it speaks the OpenAI format.
- Create a Plugsky key and test a real task against two or three models.
- Change the base URL and model names in the LLM configuration, not in the crew logic.
- Re-run a fixed task set and compare completion quality, tool calls and cost per task.
- Keep CrewAI's orchestration features and add scoped keys plus audit logging.
- Route routine crew steps to small models and reserve frontier models for planning.
Try it yourself
Open the multi-agent workflow generator →
What CrewAI gives you
CrewAI models agents as a team with roles: a researcher, a writer, a reviewer, each with a goal and an expected output. Crews run tasks in sequence or hierarchy, while flows add event-driven control with explicit state and conditional routing. Tools are ordinary Python functions, and memory can be short-term, long-term or entity-based.
The framework is deliberately opinionated, which makes simple multi-agent patterns quick to build. It is also framework code running inside your process: you host it, scale it, sandbox any code execution, and own observability and evaluation. The model layer is bring-your-own, which is the seam where a platform fits.
Framework plus runtime: the split that works
CrewAI decides how agents collaborate. Plugsky decides how model calls are served, secured and billed. Keeping that separation lets each layer do its job without a rewrite.
- Framework: roles, task graphs, delegation, memory and tool wiring.
- Runtime: model catalogue, function calling, streaming, JSON mode, embeddings.
- Interface: OpenAI-compatible HTTP, so integration is configuration.
- Governance: scoped keys, RBAC, SSO/SCIM and audit logs at the platform level.
- Evaluation: stays with you, and one key makes model A/B tests cheap.
Where CrewAI is more machinery than the task needs — a single tool-calling loop, for example — a plain loop over chat completions with your own state is easier to debug and cheaper to run.
An honest comparison
A framework and a runtime are not substitutes, and neither one replaces the other. If you want someone else to operate the agent loop entirely, a managed agent product is the category you are looking for. If you want model independence and predictable platform cost while keeping CrewAI, change the provider and keep the framework.
Plugsky serves the runtime path: 30+ models on one OpenAI-compatible key with live function calling, streaming, JSON mode, embeddings, RAG and agents, plus deployment from shared cloud to VPC, on-prem and air-gapped. Assistants, files, batch and fine-tuning endpoints are coming soon. Self-serve plans are flat monthly and the free tier covers plugsky-micro and plugsky-lite; current plans are on the live pricing page.
Honest comparison
| Need | CrewAI alone | CrewAI plus Plugsky | Managed agent product |
|---|---|---|---|
| Agent design | Roles, crews and flows | Roles, crews and flows | Vendor-defined patterns |
| Model access | Bring your own endpoint | 30+ models behind one key | Vendor's model list |
| Hosting | You run the process | You run the process | Vendor runs it |
| Governance | Your responsibility | Scoped keys, RBAC, audit logs | Vendor controls |
| Exit cost | Framework-specific code | OpenAI-compatible interface | Highest |
Frequently asked questions
Is Plugsky a drop-in CrewAI replacement?
No. Plugsky replaces the model provider and platform layer, not the framework. You keep CrewAI and point its LLM configuration at Plugsky's OpenAI-compatible endpoint.
How do I connect CrewAI to Plugsky?
Configure the OpenAI-format LLM client with the Plugsky base URL and API key, then select model names from the catalogue. Crew logic stays unchanged.
Which is more work to operate?
CrewAI requires you to host and scale the runtime and manage provider integrations. Plugsky provides the model runtime, keys, audit and deployment options, reducing the operational surface.
Do I lose CrewAI's memory features?
No. Memory, entity stores and context handling stay in your framework. You can complement them with Plugsky embeddings and RAG for retrieval-heavy workloads.
How does pricing compare?
Plugsky self-serve plans are flat monthly with unlimited fair-use usage; see the live pricing page. CrewAI is open source, but you pay for the models and infrastructure you run.
Can I use another framework instead?
Yes. The same pattern works with AutoGen, LangGraph, LangChain, Haystack, Semantic Kernel and the OpenAI Agents SDK: keep the framework, change the base URL.
What about code execution in crews?
Keep it sandboxed with no ambient credentials, exactly as you would with any model provider. Plugsky supplies the model API, not the execution sandbox.