D
Maya Chen
CaptainDirect · Results-first
Maya, bottom line: the pilot cut onboarding time 40%. Two rollout options attached. Can you pick one by Friday?
Why it works: Leads with the result, offers a decision, sets a deadline. No warm-up.
Personality API
The personality API for agents that talk to people.
People data for AI agents. Resolve anyone by email, LinkedIn URL, or name and company, and get back a validated personality profile, the playbook for that person, and your message rewritten to match. One API key, a REST API, and an MCP server your agents call mid-task.
REST APIMCP serverPay per call
The difference
Same intent, adapted to the person.
Generic AI gives everyone the same answer. An agent with Crystal context delivers the same message two different ways, because the two people reading it could not be more different.
One intent
Follow up after the pilot and move toward a rollout decision.
S
Sam Alvarez
SupporterSteady · Relationship-first
Hi Sam, thanks for the thoughtful questions last week. The pilot went well, and I wrote up how the rollout would work step by step, so nothing changes for your team overnight.
Why it works: Opens with the relationship, reassures about change, walks through steps.
What your agent gets
Four abilities, one connection.
Connect once and every workflow downstream inherits the people layer. These are tools the agent calls on its own, mid-task, whenever a person enters the picture.
- 01
Identify anyone mid-task
GET /v4/profileYour agent resolves a real person from an email address, LinkedIn URL, or name and company while it works: no manual lookup, no copy-paste, no separate tab.
- 02
Understand how they communicate
GET /v4/content/profile/:idEach profile carries a validated DISC type and archetype plus concrete communication guidance: pace, detail level, what builds trust, what creates friction.
- 03
Adapt the output, not just the data
POST /v4/content/revise_emailThe agent rewrites the email, call script, or talk track for that specific person. Same intent, different delivery, grounded in the profile instead of a guess.
- 04
Prepare for the room, not just the person
relationship_matrixPull upcoming meetings and get group guidance built from every attendee’s personality: where styles will clash, who needs detail, who needs the bottom line.
Where it runs
Built for agents that talk to people.
Sales agents
Outreach, objection handling, and follow-ups tuned to how each buyer decides.
Support
Responses that match the customer’s patience, detail tolerance, and tone.
Recruiting
Candidate outreach that lands on the first touch instead of the third.
Customer success
Renewal and expansion conversations framed for the stakeholder reading them.
Voice agents
Live calls that adjust pacing and directness to the person on the line.
Meeting follow-up
Recaps and next steps written the way each attendee actually reads them.
Why not just prompt for it
A model can guess a type. This one is checked.
Paste a LinkedIn profile into a chatbot and it returns a confident DISC type with nothing to check it against. Crystal profiles come from a prediction model trained on real assessment results, with checks that hold back a type when the evidence is thin. Graded blind, Crystal outscored all 4 frontier models on the tricky spots.
See the full benchmarkHead to headAugust 2026
How many of the tricky spots in a person’s results each profile handles
100 real, anonymized people. Same scores, same prompt, one blind judge. Higher is better.
Crystal70.2%
70.2%
GPT-5.6 Terra56.1%
56.1%
Claude Sonnet 4.647.2%
47.2%
Claude Opus 4.846.0%
46.0%
GPT-4o12.8%
12.8%
0%50%100%
For developers
An API your coding agent can wire up.
Bearer-token auth, JSON responses, async profile creation with job polling. Keep your own model in charge with the free prompt endpoint, or let Crystal return the adapted output directly. The reference ships as Swagger plus llms-full.txt, so point your coding agent at the real spec instead of a guess.
GET
/v4/profileLook up an existing profile by email, LinkedIn URL, or name and company. Re-fetching the same profile within a billing cycle is free.
POST
/v4/predictionsCreate a profile for someone new. Async job, usually done within tens of seconds. Idempotent by record_id.
GET
/v4/content/profile/:idFull personality content: DISC type, traits, and communication guidance.
GET
/v4/content/profile/:id/selling_toThe selling-to playbook for that person: how they decide, what wins trust, what to avoid.
POST
/v4/content/profiles/relationship_matrixGroup guidance across a set of profiles: where styles clash and who needs what.
POST
/v4/content/revise_emailSend a drafted email, get it rewritten for the reader with a list of what changed.
POST
/v4/content/generate_promptA personality-tuned prompt for your own model. Bring your model, keep your stack. Free, no credit spent.
Access and pricing
Three ways in, pay for what you call.
The Personality API is built on the same validated profiles that power the rest of Crystal. Sign up on the data dashboard for a key; larger volumes and enterprise terms go through a demo.
REST API
PreviewBearer-token auth, JSON responses, 120 requests a minute per token. Swagger reference plus llms.txt and llms-full.txt so your coding agent can wire it up from the real spec.
Crystal MCP server
PreviewConnect Crystal to Claude, ChatGPT, Copilot, and any MCP-compatible agent framework. Profiles, guidance, and meeting prep become tools your agent calls mid-task.
Claude plugin
PreviewOfficial open-source plugin for Claude Code with OAuth sign-in and two ready-made skills: pre-meeting prep and call transcript analysis.
View on GitHub
How credits work
- API credits are spent when a new profile is created or an email is revised. Failed requests cost nothing.
- Fetch credits are spent on reads. Fetching the same profile again in the same billing cycle is free.
- The prompt endpoint is free and spends no credit at all.
Give your agents people skills.
One connection brings personality profiles, communication guidance, and adaptive output into every workflow you run.