Written for heads of product and platform at AI writing, sales engagement, CRM, social and agency software companies.
What a licence includes
Each item below is live today and documented on its own page.
REST API
Fetch a workspace's context, draft from it, review a draft and record approvals. Bearer keys with scopes; nothing in it can publish. API reference
Remote MCP server
The workspace context arrives as resources when an assistant connects. Prompts cover a post, a weekly plan and a cold email. MCP server
Context gaps
When a draft names a claim the saved facts do not cover, the draft is flagged, so the gap is visible before anyone posts.
Voice from real posts Beta
Voice is measured from the user's own posts: sentence length, fragments, punctuation and contractions. Drafts are held to those numbers.
Today's plan for agents
A read-only view of the GTM work due today and over the next 7 days, as a tool and as a resource an agent can read.
What the test showed
With the same model on every brief, drafts written with Klype’s context introduced 0.08 names not in the facts or brief on average, against 22.75 for the model alone, and no unsupported figures, against 0.67.
| Average per draft | Model alone | With Klype context |
|---|---|---|
| Names not in facts or briefProduct, company or tool names the draft introduces | 22.75 | 0.08 |
| Distance from measured voiceDifference in sentence length, spread, fragment rate and contractions | 0.52 | 0.18 |
| Banned AI phrasesA fixed list, such as delve, unlock and seamless | 1.08 | 0 |
| Unsupported figuresPercentages, money, durations and counts not in the facts or brief | 0.67 | 0 |
| Invented product claimsSentences about the product whose words are mostly absent from the facts and brief | 1.58 | 0.50 |
| Trap feature termsWords stating a feature the product does not have, on the two trap briefs | 0.17 | 0 |
| Review score, higher is betterScored by Klype’s own checker, so read it as self-reported | 65 | 83 |
Source: Klype’s context evaluation, 30 October 2026, fictional company. Method below.
Both trap briefs, which assert features the product does not have, were flagged by Klype’s gap check. That check is Klype’s own, so it is self-reported.
Where Klype was not better
- 3 of the 12 briefs.
- On the how-to and numbers posts, invented product claims were 0 against 2 on one brief and 1 against 2 on the other. The extra claims were generic lines, not false features.
- An 11-word X post scored further from the measured voice.
Method
Klype’s context evaluation, run on 30 October 2026. Both columns use the same model, DeepSeek V3.2, on 12 fixed briefs for a fictional company. Every brief is reported. For every row except the review score, lower is better.
This is one run, with one model, on one fictional company. It says nothing about any real customer.
Why it gets stronger with each model release
Same context, better model
A better model uses the same saved facts, voice and plan better. The context does not change when the model underneath does.
Every MCP assistant is a new place
Any assistant that speaks MCP can connect to the same server, so each new one is another place Klype plugs in.
The context is the customer’s own
It belongs to the customer, works across vendors and is shared with their team, so it does not sit inside one tool.
It compounds with use
As facts are saved, drafts approved and results imported, the workspace grows, and each draft reads the current version.
Pilot
We run the same with-and-without test on your model and your briefs. You see every draft and score before any commercial talk.