prism-2Default
Our most accurate model, for research and writing with citations.
- Context
- 1M
- First token
- 380 ms
- Output
- 95 t/s
Product
One assistant for everything your team knows.
Search across every tool, write with citations, put recurring research on a schedule and analyse files, with the security controls your IT team will ask about.
01 · Search
Ask once. Prism searches everywhere.
Prism plans a search across your drives, wiki, help desk, CRM and the web, reads the passages that matter and ranks them before it writes a word.
- Hybrid keyword and semantic search, tuned per source
- Filters for date, author, source and file type in plain language
- Flags when two sources disagree instead of picking one
02 · Write
Drafts with the receipts attached.
Turn research into briefs, emails, reports and specs. Citations stay attached as you edit, and Prism re-checks them when you change a sentence.
- Follows your team’s style guide and templates
- Unsupported sentences are underlined, never hidden
- Export to docs, slides or Markdown with footnotes
03 · Briefs
Recurring research, on a schedule.
Save any question as a brief. Prism re-runs it daily or weekly, highlights what changed since last time and posts the result to email, chat or a webhook.
04 · Analyse
Spreadsheets, PDFs and exports, explained.
Upload a CSV, a 300-page PDF or a warehouse table. Prism runs the numbers in a sandbox, shows the working and charts the result.
Integrations
Connects to where work already lives.
40+ connectors sync every few minutes and keep source permissions. Anything else can push documents through the API.
- Cloud drivesDocs, sheets, PDFs
- Team wikiPages and databases
- Team chatPublic channels
- Help deskTickets and macros
- CRMAccounts and deals
- Git repositoriesCode, issues, PRs
- Data warehouseRead-only SQL
- EmailShared inboxes
- CalendarMeetings and notes
- Web searchWith domain filters
- PDFs and scansOCR included
- Your ownDocuments API
Security and privacy
Your data trains nothing. Ever.
Customer content is never used to train models, on any plan. Data is encrypted in transit and at rest, stored in the region you choose and deleted on your schedule.
- SOC 2 Type IIReport on request
- ISO 27001Aligned controls
- GDPRDPA and EU hosting
- SSO and SCIMSAML 2.0, OIDC
- AES-256 at rest, TLS 1.3 in transit
- Audit log of every query and export
- US, EU or Canada data residency
- Annual third-party penetration tests
Data retention
How long Prism keeps each kind of data, by plan.
| Data | Free | Team | Enterprise |
|---|---|---|---|
| Chat history | 90 days | You choose | You choose |
| API requests | 30 days | 30 days | Zero retention |
| Source index | While connected | While connected | While connected |
| Audit logs | — | 1 year | 7 years |
| Used for training | Never | Never | Never |
Need our security questionnaire or SOC 2 report? Request it from sales.
Models
Fast where it matters, careful everywhere.
Three models, one API. Prism picks the right one in the app; through the API you choose.
prism-2-flash
Low latency for search boxes, chat widgets and high-volume jobs.
- Context
- 256k
- First token
- 140 ms
- Output
- 240 t/s
prism-embed-1
Embeddings for your own retrieval, tuned on the same data as search.
- Input
- 8k
- Dimensions
- 1,024
- Latency
- 25 ms
Milliseconds, measured from our US-East region with a 2,000-token prompt.
See Prism on your own sources.
Connect a drive and a help desk in five minutes. Free for individuals, 14-day trial for teams.