One platform
Instead of ten services that have to be wired to one another — a single system with shared permissions and shared accounting. There is nothing to integrate: inside, it is already connected.
Enterprise AI platform · August 2026
One platform instead of a dozen AI tools
Any model, spend visible per employee, agents governed by an administrator and the company's shared context — in one system. It runs in our cloud, and where data must not leave the company, inside its own perimeter.
01
Problem
Companies buy AI piece by piece — and lose sight of it
The tools arrive one at a time: a department pays for one vendor, engineering works with another vendor's keys, support runs a bot nobody remembers. Nobody holds the whole picture — and that is not a matter of discipline, it follows from there being no shared system at all.
01
More tools than anyone can hold in their head
Every new tool means a separate invoice, separate permissions and a separate place data can leak from. None of them knows the others exist.
27
AI applications in the average company; their share of the software portfolio
grew from 7% to 22% in a year
BetterCloud, 08.2026
02
There is a policy on AI spend — there is no instrument that shows it
The invoice arrives as a single number: no way to see who spent it, on what, or when. What you cannot see, you cannot manage — you can only be surprised by the total.
73% / 13%
have a policy on AI spend — and this many can actually see the spend. Around 26%
of the budget is wasted
Harness, 07.2026
03
Agents appear on their own, outside any control
Someone spins up an agent in an outside service, puts it on a schedule — and every morning it reads something and spends money. The person leaves the company; their agent keeps working.
82%
of organizations found agents they did not know about. 91% use agents; 10% govern
them maturely
Cloud Security Alliance · Okta, 2026
04
The company's context sits nowhere — and models cannot see it
Why this decision was taken, who objected, what had already been tried and dropped — all of it is scattered across tickets, chats and documents. Every answer has to be reassembled by walking round colleagues, and when a person leaves, the context leaves with them.
124,750
person-to-person links in a company of 500. Queries to a platform that holds that
context: 500
n(n−1)/2 vs n
02
Solution
One system instead of a set of disconnected tools
The company connects its own model providers and starts working — in our cloud right away, or inside its own perimeter if its rules demand it. Below are the three things the platform actually gets installed for.
Agents under control
An employee creates the agent; the administrator sets the frame — which tools and models it may use, what its spending ceiling is, whether it runs at all. The agent sees exactly what its creator sees, and that is enforced by the engine, not by the wording of the task.
Company context
Everything the company knows is gathered in one place and stays there when a person leaves. People and agents both draw on it: a new agent does not have to be told how things work here — it already knows.
-
01
One door to every model
Anthropic, OpenAI, Google, xAI — or a local model on your own hardware. The company's keys, the company's rules, one place where a model is switched on and assigned to a job.
-
02
Spend visible per employee
Spend by person, by model and by agent, with a weekly ceiling per person and an alert when a threshold is crossed. Chat and agent work are counted separately, so you see not only how much, but on what.
-
03
Agents created by employees, governed by the administrator
The administrator sets the ground rules agents are built on: which tools they can reach, which models they can be given, what the spending ceiling is. From there they see every agent in the company and can stop any of them; disabling an employee stops their agents too.
-
04
Company knowledge in one store
Connectors pull Confluence, Jira, GitLab and Slack into a single store with hybrid search and a graph of relations. Source-system permissions are honored per document: a person sees in the answer exactly what they see in Jira itself.
-
05
One way in from anywhere
Web chat, Slack, Telegram, Mattermost, MCP and an external API — six entry points on a single core. People work where they already are.
-
06
Useful from day one
The system is useful immediately as a shared door to models with transparent spend, even while the knowledge base is empty, and matures into company memory as sources are connected. There is no mode to switch: the first document loaded turns search on by itself.
03
Product
The very instance you can open right now
Below are screenshots from the live instance, not mockups. Access is open: the login and password are printed above, and no signup is needed.
04
Status
Not a concept — a system that works
Version one is complete and verified end to end: from pulling data out of the sources to an answer arriving in a messenger. The platform runs on a public instance you can log into right now.
- v1
- complete end to end and verified live
- 1,100+
- automated tests across the system
- 6
- entry points on a single core
- 4
- connectors to knowledge sources
- 1
- command to bring the whole system up
Python 3.14
FastAPI
React · TypeScript
PostgreSQL · pgvector
Redis · SAQ
Docker Compose
Nginx
AGPL-3.0
The knowledge graph, the vectors and the permissions all live in one PostgreSQL — no separate graph engine, no external vector database. Hence cheap operation: one system rather than a zoo of services to keep alive. And hence a capability few can offer — the same platform moves into a client's own perimeter whole, down to a network with no internet at all.
05
Timing
Companies need a platform, not one more tool
01
The mid-market buys these systems rather than building them
A year ago half of them were assembled in-house; now three quarters are bought ready-made. A company of two or three hundred people has no team to glue together a model gateway, a vector store, permissions and spend accounting on its own.
76%
of production AI solutions are bought rather than built in-house — against 53% a
year earlier
Menlo Ventures, 12.2025
02
The market has grown large enough to start counting money
Analysts record the shift in what is being asked: from "give us access to models" to "show the return and keep the spend under control". That is precisely the job this platform was built for.
$64B
the market for AI platforms and models in 2026, up 63% in a year
Gartner, 07.2026
03
Part of the production work is leaving the public cloud
The turn took a single year and concerns production work, not experiments. We cover both sides: an ordinary company is well served by our cloud, and those who cannot let data out get the same system inside.
56% → 41%
public cloud as the primary environment for production inference. 56% of
companies use or plan private cloud
Broadcom, 06.2026
04
For some buyers this is a question of admission, not taste
The AI Act and Data Act in Europe, MAS requirements in Singapore, data localization in Vietnam, Indonesia and India. A bank or a hospital cannot buy a solution that keeps their data in someone else's cloud — which puts them out of reach for most competitors.
90%
of organizations consider local data storage more secure; 64% fear leaks through
AI tools
Cisco, Data Privacy Benchmark
06
Market
Who buys this, and for what money
The buyer is a company of one hundred to a thousand people that already runs several AI tools and has no platform team to connect them. The decision sits with the IT director: 65% of them hold a dedicated AI budget in 2026, against 36% two years earlier. A separate segment is organizations required by a regulator to keep data inside their own perimeter.
TAM
up to $31B
The intersection of the categories the platform works in: LLM operations,
enterprise search and knowledge management.
LLMOps $7.1B · Enterprise search $7.5B · Knowledge management $16.2B · 2026
SAM
$2.8–3.4B
Our two channels together: the mid-market in the cloud, and the segment obliged to
keep data inside. Both shares grow faster than the market as a whole: 12% and 19% a
year against 9%.
Estimate: on-premise ≈33% and mid-market ≈28% shares taken from an adjacent
category — an assumption, not a measurement
SOM
≈ $14M ARR
The reachable share at the price of the closest open alternative: two hundred
companies of three hundred seats each.
200 companies × 300 seats × $240 a year
The pace of the category shows in its leader: Glean went from $200M to $300M in annual revenue in five months and is valued at $7.2B. In August 2026 it described the market's problem in the same words we do: gaps in context, rising costs and AI sprawl across the company.
07
Model
Subscription for most, a license for the large
The main path is a subscription to our cloud: the company brings nothing up itself and pays for each working employee. Large and regulated organizations that need the system inside their own perimeter buy an annual license.
| What we sell | Price | For whom |
|---|---|---|
| Cloud · Team | $15–19 per employee / month | Smaller companies: ready to work, nothing to deploy |
| Cloud · Business | $24–29 per employee / month | The mid-market: higher limits, every connector, support |
| Own perimeter · Enterprise | $25–150K a year | Large and regulated. The price follows company size: around $25K for a hundred employees, around $150K for a thousand. On top of that — audit, retention, an isolated perimeter, warranties and a service level agreement |
| Tokens and storage | usage-based | A pool comes with the subscription; beyond it, a transparent rate |
Price as a difference. The market median is $24 per employee per month; the closest open competitor charges $20, the Microsoft add-on $30. Meanwhile neither Glean nor Notion publishes a unit price for consumption — a buyer cannot work out their own bill in advance. We can give them that, and in this category it is rare.
08
Roadmap
From a knowledge base to company memory
Search knows what exists. The graph knows what is connected to what. Beyond that begins something nobody does today: understanding why the company ended up the way it did.
-
Now
A library of ready-made agents
Agents you switch on instead of writing: company analytics, digests, watching over processes. The first set is the company looking at itself — where a service rests on a single person, where two teams do the same work, where documentation is missing.
-
Next
The layer of reasons
Not only documents and links, but the history of decisions: who proposed, who objected, which alternatives were weighed and why they were dropped. That is what turns a knowledge base into the company's memory.
-
A product of its own
Development on company context
The platform already holds everything the development process rests on: tickets, code, discussions and the reasons behind past decisions. A separate offering grows naturally on top — agents carrying a task from statement to finished change, by the company's own rules. A second product on the same foundation, with a price of its own.
-
Horizon
Agents that act ahead of you
A personal assistant holding the company's full context notices a risk by itself, surfaces the relevant past experience and warns a person before they run into the problem. Beyond that — agents that carry out whole tasks on company knowledge and by company rules.
09
What we look for
The product is built. Code is not what is missing
What is missing is proof of demand: live deployments that pay, and a team to serve them. We are looking for a partner — a fund or a company — to back this stretch of the road and help walk it. The amount and the terms are a separate conversation.
-
01
Three to five paying deployments
Not pilots for show, but companies that keep the system in daily work. They, and not the popularity of a repository, are what shows the product is needed.
-
02
A co-founder and the first engineers
The product is ready, but a product is not a company. We need someone to take on go-to-market, and a small team to sustain the pace of supporting clients.
-
03
Pricing tested on real deals
The price range is derived from market benchmarks but untested in negotiation. It needs to meet real buyers and be corrected by what comes back.
-
04
A way into the first market
Southeast Asia and Europe look like the natural entry: there, data residency requirements make working inside your own perimeter not an advantage but a condition of admission — and they cut most competitors out.
Open it and see for yourself
The instance runs continuously, on a live knowledge base. No signup — create an agent, ask a question against company knowledge, and see what it cost.
Email
semvitphone@gmail.com
Telegram
@DemionRock