PrymeIntelligence
The Governed Agentic OS

Build agents you can prove — and own.

Certified before deployment
Every agent passes governance gates with a tamper-evident evidence ledger.
You own your memory
Context Protocol keeps your agents' memory yours — portable and sovereign.
Build in chat, deploy anywhere
No code required. Web, mobile, WhatsApp, email, or API — one workspace.
SOC 2 · ISO 27001-aligned · GDPR-aware · trusted by serious teams
Back to prymeintelligence.com
Welcome back
Sign in to your workspace.
OR
No workspace yet? Create one free.
Your organisation signs in through its own identity provider? Use single sign-on.
Back
Check your email
Factor one · the code we emailed to your work email · good for 10 minutes
Didn’t get it? Send it again
Back
Second factor
6-digit code from your authenticator app
Lost your device? Use a recovery code
The emailed code proves you hold the address. The authenticator proves you hold the device. Both are required on every sign-in to a workspace that can act on live systems — there is no remember-this-device.
Step 1 of 4
Back to prymeintelligence.com
Create your workspace
Start free. No credit card required.
OR
We send a code here before anything else is set up.
Already have a workspace? Sign in
Step 2 of 4
Change email
Check your email
Code sent to you@company.com · good for 10 minutes
Didn't get it? Send it again
Step 3 of 4
Email verified
Add your second factor
The emailed code was factor one. Add this workspace to your authenticator app with the setup key below, then enter the 6-digit code it shows.
Your recovery codes
Ten codes, shown once. Each signs you in once if you lose the device. Keep them where you keep passwords — they are not emailed and cannot be shown again.
Step 4 of 4
Back
Name your workspace
We suggest a name from your email domain — change it to the name you trade under. Three things, and you are in.
Read from your email domain. Change it if it is not the name you trade under.
Decides the region your ledger, knowledge and vectors are written in. The United Kingdom (London) and the United States (Iowa) are regions of their own; elsewhere is written in the nearest one we run, and the workspace says which.
Install Pryme on this computer.It opens from your Dock like any app, signed in.InstallNot now
Overview
Search agents, runs, evidence…⌘K
synced —
New agent
Today
now
00:00
— evals todayallowed —refused —held —Audit Trail
at or above its floor · passes70 up to its floor · below floor< 70 · failsno evidence yet
Attest · where depth is earned
Add knowledge
What it is decides how an agent may use it. How it gets in decides how it stays current.
What kind of knowledge is this
Which agents may read it
⚲
Ask as an agentlexical retrieval · no vector index
Put a question to what one agent may read, and see what it would cite — or why it would refuse. This is what the agent itself gets: passages with citations, never a guess. The question is logged here as a masked pattern; it never leaves this workspace.
Coming soon
This module is part of the Pryme roadmap.
In your Phase 1 workspace
Attest and Context Protocol ship first. Agent Studio, Knowledge Base, and the wider governance suite roll out next — you'll see them light up here.
Bring an agent to Attest
Agents inside your own estate are brought in by engagement. Connect it, declare its role, hand over the standard it must meet.
1Source
2Connect
3Role & scope
4Standard
5Team
6Register
Where does this agent live?
Connect an existing agent
Point Attest at an endpoint or MCP server it already runs on.
MCP-native
Import from a framework
Bring an agent from the stack you already run — connected through the Pryme SDK.
Build it inside Pryme
Born instrumented in Agent Studio — certifiable from day one.
Connect the agent
Test connection
Declare its role & scope
Web chatWhatsAppEmailVoiceSlack
Connect the inbox this agent trains on & replies from
Gmail
Outlook
IMAP / SMTP
Tool access defines the blast radius Attest gates against — e.g. refund-api moves money, which is depth 4, one depth beyond this role.
Hand over the standard it must meet
Seed from Context Protocol Preview · coming
gold answers from your resolved cases, hash-referenced
Upload gold examples
CSV / transcripts of exemplary responses
＋
Upload scenario packs
.csv · .json — real cases the agent will be drilled on, case by case
Each gate clears its own floor: Grounded 88 · Refuses 92 · Depth 92 · Escalates 87 · Scope 88 · Adversarial 88 · Stable 86. Floors are the kernel’s, not a setting.
Assign the review team
Full accessManage the agent, its standard & team
Runs simulationsTrigger sandbox & benchmark runs
Provides feedbackDraft corrective-feedback proposals
Approves feedbackApprovers who clear proposals into memory
Dual · 2 approvalsSingle approval
High-risk agents default to dual review — a proposal needs two distinct approvers before it affects live responses.
Review & register
AgentCustomer Support Agent · Customer Support Agent
Source—
ScopeWeb chat, WhatsApp · 3 tools
Team—
StandardContext Protocol · gold answers (coming) · 3 policies
Certification flooreach gate its own · 7 gates
Step 1 of 6 · choose a source
Continue
P
P
Pryme · Attest
Certificate
Certified
This attests that the agent
Customer Support Agent
Customer Support Agent · certified to depth 3, its role ceiling, on web chat & WhatsApp
model anthropic:claude · sonnet prompt v12 cert — grade A
Readiness · 90 · the lowest gate, never an average
Grounded · floor 8893
Refuses · floor 9294
Depth · floor 9295
Escalates · floor 8790
Scope · floor 8892
Adversarial · floor 8891
Stable · floor 8690

Each gate clears its own floor. Grounded 88 · Refuses 92 · Depth 92 · Escalates 87 · Scope 88 · Adversarial 88 · Stable 86.

Certified per channel — not per agent
Web chatdepth 3 · form gates 4 of 4certified
WhatsAppdepth 3 · form gates 4 of 4certified
Emailnot qualified · no runholds nothing
Phonenot qualified · no runholds nothing

A right answer in the wrong form fails the channel, so each is drilled and certified on its own. Asking this agent to take a call it has not qualified for is the same refusal as asking it to act above its depth.

Certificate · 90 days · then it lapses
●In force For 90 days from issue, at the depth it earned.
■Lapsed At 90 days it lapses, and the agent drops to depth 1 at once, with the reason on the ledger.
■Failed run A failed run drops the agent to depth 1 at once, with the reason on the ledger.

Your agent runs the stack this certificate was issued against. A new model version does not touch it — the changed stack is qualified as a candidate alongside it and replaces it only after clearing every gate at the same depth.

Provenance · hash-chained ledger
Ledger height— · every entry carries the hash of the one before
Certificate hash—
Chainan altered or removed entry breaks verification from that point on
Attestation
—
closed the run · signed with the workspace’s key
—
Workspace owner
Scan to verifyIndependently confirm this certificate · —
Issued 14 Jul 2026 · lapses 12 Oct 2026 (90 days) · a change to the model, prompt or policy qualifies a candidate — it does not revoke this one.
Scope-bound: valid only for the role, model, version and channels above.
Share
Download
Done