AI Vendor Lock-In Reviews
Can you export your context from ChatGPT via API? No.
From Claude? No.
There is such a strong demand for AI that the classic requirements of enterprise software have been quietly put aside.
ChatGPT started consumer and didn't care for long about enterprise requirements. Even though Claude has been more enterprise-centered, their rise is so fast that they didn't put more effort.
Why should it matter now?
A few reasons:
Ownership. It should be clear by now that AI is here to stay and will be transformative. So companies need to ask themselves whether they want their intelligence to be their own, or whether they want to delegate it. For better or for worse.
Switch to agents. AI assistants are quickly becoming agentic platforms. It wasn't much of an issue to have a locked AI assistant, but closed agent platforms will mean your AI provider locks and controls all your workflows.
Pricing. AI is currently highly subsidised because of the race to the top, but in a year or two, AI vendors will probably raise prices on their locked-in customers.
What is it?
This article is an attempt to review every major hosted AI assistant platform to clarify what's really enterprise-friendly / agent-friendly:
Can you use or export the context the platform learned about you, programmatically?
Can an agent you build interact with it?
Can you read conversations created on the platform?
We score five platforms — Claude.ai, ChatGPT, Gemini, Microsoft Copilot, Le Chat — across four axes:
User data accessibility — what you've put in, can you get back out?
Integrations accessibility — can the platform use your tools?
Agent platform — can you build and run other agents on top of the platform?
Model openness — can you use your own model on the platform?
Hint: the results are not pretty.
A few funny things:
Google invented A2A and then didn't ship it in gemini.google.com.
Anthropic invented MCP and then didn't expose Claude.ai over MCP.
Microsoft invented neither protocol, then shipped both end-to-end.
Interactive scorecard
AI Vendor Lock-In Benchmark
Claude.aiAnthropic | ChatGPTOpenAI | GeminiGoogle | M365 CopilotMicrosoft | Le ChatMistral | |
|---|---|---|---|---|---|
AUser data opennessWhat you've put in, can you get back out? | |||||
A2Context export & portability Memories, projects, custom GPTs — re-importable? | |||||
CAgentic exposureCan your agents use it? | |||||
C1Platform exposed as MCP server Can external agents read your data via MCP? | |||||
C2Agent-to-agent (A2A) protocol Callable as an A2A agent by orchestrators? | |||||
C4Connection composition Can an MCP server reuse the platform's other auth'd tools? | |||||
C5Reverse LLM access (MCP sampling) Can a connected tool call back to the platform's LLM? | |||||
DModel opennessCan you choose the model? | |||||
D2BYO LLM at runtime ("electricity test") Paste an LLM endpoint URL — done? | |||||
Total/ 18 | 2 | 2 | 3 | 9 | 4 |
Methodology and the full repo
We've published the entire review on GitHub:
👉 github.com/myriade-ai/ai-vendor-lockin
Note: this review was evaluated, with sources, by Claude — then improved and corrected by humans.
If a score is wrong or out of date, open an issue with a source link and we'll check.