A Solidalab product
Solidai brings AI assistants, an enterprise knowledge base, visual workflows and tool-connected agents together on one platform. Every component — the language model included — runs on your own servers.
In the enterprise the question is no longer "should we use AI?" but "where is our data going?". Customer records, contracts, financial reports and internal correspondence handed to a cloud service are, in finance, healthcare, government and defence, either against regulation or an unacceptable risk. Solidai was designed as a direct answer to that problem.
Early access in preparation · MMXXVI
At runtime no query, document or log record ever leaves your network.
Application, flow engine, vector store and local model inference ship as a single install.
A pilot runs on one machine and grows on the same architecture as demand rises.
An end-to-end platform, from chat assistants grounded in your internal knowledge base to a visual flow designer that automates business processes. All five components ship inside the same installation, and none of them calls out to an external service.
Assistants grounded in your own documents and data, configured for a department, a team or a single user. Role, knowledge base, available tools and tone of voice are all set from the admin panel — without writing a line of code.
PDF, Word and text files are uploaded, split into meaningful chunks, embedded locally and indexed. The assistant searches that base while answering and cites the source it used, so replies rest on your own documents rather than guesswork.
Multi-step automations are built on a drag-and-drop canvas: an assistant answers the incoming request, the result passes a decision node, a database query runs if needed and the output becomes a report. Every step is visible and can be tested on its own.
AI that acts, not just talks. Define agents that query your database, reach the file system and speak to your internal services. Each agent touches only the resources it was explicitly granted — the boundary is enforced by the platform, not left to the model's discretion.
Every conversation, tool call and flow step is written down with a timestamp. "Where did this answer come from?" is answered by the ledger: the model used, the source retrieved, the tool executed. Records can be exported for audit teams.
A request arrives, the assistant searches the knowledge base, the result passes a decision node, a query hits the internal database if required and the output becomes a report. Every step runs on the same server, inside the same wall.
One installation, independent assistants. Each unit raises its own arch: its own knowledge base, its own tools, its own permission boundary. The arches carry their load separately, yet all of them stand on the same stylobate — one platform, one wall, one ledger.
Personnel files, policies and procedures; answers employee questions with the source.
Clause search, comparison and summarisation across the contract archive.
Reports and reconciliation records, with agents issuing permitted database queries.
Customer requests: answers grounded in product docs, escalation by workflow.
Field procedures and maintenance instructions; immediate answers on shift.
One unit's documents are invisible to another unit's assistant; the boundary is enforced by the platform on every query. Adding a unit is simply laying one more arch in the arcade — no new server, no new licence stack, no new installation.
Solidai installs as a single stack and runs entirely inside your network. Its layers are arranged like the concentric rings of an arch: each outer ring protects the one within, and at the very centre sits the language model running on your own hardware.
The user interface, admin panel, authentication and authorisation. Who may reach which assistant and which data is defined here.
Assistants, workflows and tool calls execute in this layer, with permission boundaries re-checked on every call.
Vector store and primary database: documents, embeddings, conversation history and audit records all stay on the same server.
Inference runs on your own hardware. Neither prompt nor response leaves the server, and the model weights stay with you.
There is no outward dependency; the installation can be completed even in fully air-gapped environments.
Each line is an event: which assistant ran, which source was retrieved, which tool was permitted and which was refused. An audit question is answered by the record, not by recollection.
Solidai was designed to run without any cloud connection — even in an environment completely severed from the internet. Language model, vector database and primary database all run on your servers.
This is not a "your data is encrypted in transit" assurance: no outbound call is ever made. The difference is enforced by architecture, not by contract.
Creating an assistant, attaching a knowledge base or designing a workflow needs no developer. Business units use the platform directly for their own needs; the technical team is involved only at installation and integration.
Who can reach which assistant, which data and which tool is set at a fine grain. Users, roles and access are managed from a single panel, and every access is written to the audit ledger.
Built to start as a pilot on one server and expand on the same architecture as demand grows. There is no cloud bill that scales with usage, no data transfer exposure and no third-party dependency.
Four phases, from installation to the first assistant — built the way an arch is built: first the measure, then the piers, then the stones, and last of all the keystone.
One stack installs on your own server: application, database, vector store and model inference. Hardware and residency stay entirely under your control.
Corporate documents are loaded into knowledge bases: chunked, embedded locally and indexed for fast retrieval. Access boundaries are defined at this stage.
Assistants, knowledge bases, tools and workflows are wired together from the panel, each one tested before it goes live.
Teams start using it and every step is recorded. Real usage data is then used to refine assistants and flows over time.
Built for organisations that must keep their data private yet do not want to give up the productivity of AI. When the material is customer records, contracts, patient files or classified documents, where the processing happens is not a preference but an obligation.
Obligations under KVKK and GDPR can be met without handing data to a third-party processor, and data-residency requirements are satisfied by the deployment itself.
"An arch does not carry its load away;
it distributes it within itself."
Solidai is in early access. Join the list and you will hear first when the release is ready; if you would rather discuss your organisation's needs directly, book a short introduction call.
Used only for early-access and launch announcements. Nothing else, shared with no one.