The same protocol, the opposite premise.
Crystal is the better tool if you need to understand someone you have never met; Throughline is the better tool if you need to work better with the colleagues you see every day. Both connect to Claude and ChatGPT over MCP, so the connector is not the difference. Crystal looks up one person at a time; we are the first to put a whole team map — and the guidance for any two colleagues in it — behind that connection.
- Accurate as of 23 August 2026
- Their wins listed first
Crystal is a serious product and has been at this considerably longer than we have. They shipped an MCP server before we did, their content estate is far larger, and on their core use case they are simply better than us. This page is written on the assumption you will check.
Crystal predicts profiles from public signals, which gives it coverage of millions of people who never opted in. Throughline only holds profiles from an assessment the person chose to take, which gives it consent, mutuality and a measured relationship layer — and no coverage at all outside your own team.
What they do better than us.
Coverage, and it is not close
If the person you need to understand has not taken an assessment — a prospect, a candidate, a counterparty — we have nothing and they have something. This is the whole game for a sales team, and we are not competitive on it.
Framework breadth
Six frameworks against our one. If your organisation already runs on DISC or MBTI vocabulary and wants that preserved, they meet you where you are.
Maturity
More integrations, a longer track record and a published accuracy benchmark. We are earlier than they are and it shows in the places you would expect.
Sales workflow
Prospect lists, meeting prep, CRM connectors. That is a real product built for a real job, and we do not have an equivalent.
The differences that actually decide it.
Observed from Crystal’s own public material as of 23 August 2026. We repeat no pricing or statistic we haven’t verified directly.
| Crystal | Throughline | |
|---|---|---|
| Core premise | Know the person across the table before you meet them | Work better with the person you see every day |
| Where profiles come from | Predicted from public signals, plus an optional self-assessment | Only an assessment the person chose to take |
| Frameworks | Six, including DISC, MBTI and Enneagram | Big Five, presented as four working styles |
| Coverage | Millions of people, most of whom never opted in | Only colleagues who opted in |
| Consent | The subject often does not know a profile exists | Explicit, mutual, and revocable at any time |
| Direction of visibility | One-way — you see them, they learn nothing about you | Mutual — you see them only if they can see you |
| Relationship layer | Look up two people and compare by eye | The pair is measured in its own right, with its own guidance |
| AI integration | MCP server with profile lookup tools | MCP server built as the primary product surface |
| Refusals | Few documented limits on what it will answer | Ranking, hiring, promotion and anything unmeasured are absent from the schema |
What we do that they don’t.
Consent as architecture
No profile exists without the person choosing to create it, and sharing runs both directions. That is not a setting — there is no code path that produces a profile any other way.
measured, not predicted
A prediction from public writing is an inference about someone. An assessment is a statement by them. When the guidance is about how to give a colleague difficult feedback, the difference matters.
The pair layer
Guidance about the relationship between two named colleagues, measured in its own right. Not two profiles handed over with the synthesis left to you.
Enforced refusals
There is no endpoint that ranks people or supports a hiring decision. An assistant asked to do it declines and explains why, rather than obliging carefully.
Which one you should actually buy.
Choose Crystal if
- You need to research people outside your company
- Your primary use case is sales prospecting, or screening candidates before they join
- You want six frameworks rather than one
- You need a mature product with a long track record
Choose Throughline if
- The problem is a conversation with a colleague
- Your team already works daily in Claude or ChatGPT
- Consent needs to survive a works council or a legal review
- You want the pair, not just two profiles
Ask us the question you'd ask them.
Twenty minutes, live, on a demo team. Bring the thing you think we'll be worse at — that's the useful part of the call.
See pricingTell us what you’re working on.