What exactly is Doctiling?

A knowledge management system. Your information lives in knowledge bases you govern — one per client, team or case — with documents, tables and graphs, its own system prompt, its own tools and its own permissions. Your experts curate that content and AI agents analyse it, compare it and turn it into tables, charts, diagrams and code, always citing where each claim came from. The base is shared with your team, published or connected over MCP to the tools you already use.

Is it a document manager or a vector search engine?

Neither, although it does what both do. It stores your sources with permissions and indexes them by meaning, but that is where its work starts: structuring them into documents, tables and graphs, letting your experts curate them — edit, validate, retire what is stale — and having AI agents analyse them with citations. What you get is not a file found, but verified knowledge you can reuse, share and query from other tools.

How is this different from a generic AI chat?

Generic chats answer from memory and sometimes invent. Doctiling indexes every block of your documents and retrieves the relevant passages before answering, so every claim ships with a citation back to your own content. If it's not in your sources, the agent says so instead of guessing.

Does this replace the RAG project we were going to build in-house?

For most cases, yes. Chunking, embedding, the vector index and citation are already built and tested; you need no pipeline of your own and nobody to maintain it. What stays in your hands is what actually decides the outcome: which sources enter each base, who has access, which tools the agent may call and which systems query it. If your case demands full control of the stack, the self-hosted build runs the same product on your infrastructure.

Who can see my documents?

Only you and the people you explicitly invite, under roles (owner, admin, collaborator, reader). On the Free plan your data is logically isolated per user; Gold moves you to your own dedicated database, and Platinum adds dedicated infrastructure on top.

How do I get free access?

Free access is sponsored and invitation-only. Request a demo with your email; once we approve it you can sign in and use the full studio — knowledge bases, document editor, AI agent with citations, mirror view, custom tools — with a sponsored AI budget and a quota of 25 AI calls per day. No credit card, no trial clock. Upgrade only if you need isolation or higher volume.

What is MCP and why should I care?

Model Context Protocol is an open standard that lets external AI tools read from a data source. With Doctiling you can expose a knowledge base through a scoped, revocable, read-only token, so the desktop AI tools your team already uses can query your documents as a source of truth.

Does it work in Spanish?

Yes — the whole studio is bilingual (Spanish and English) and you can switch at any moment. The agent reads and answers in the language of your documents and your questions.

Doctiling

EN ES

doctiling · intelligent processes with knowledge bases and AI agents

You don't need a system for every process.

You need organised knowledge and an agent that sets it in motion.

today

Today, automating a process usually means building screens, data models, permissions and integrations, and maintaining one more system. Or relying on email and spreadsheets.

with doctiling

Doctiling proposes another way: organise the information in knowledge bases with their own access rules and connect an AI agent that can query it, relate data and run authorised actions in plain language.

Fewer systems to build. More processes solved.

Sponsored free access · by invitation · no credit card

What is a knowledge base?See real captures Install it free on your machine

Base · Law firmcases + contracts · text + table

Check whether we have pending cases.

search_knowledge_base(Cases) · aggregate_rows(Cases, status = pending)

3 pending cases: two with a hearing this month (Andina Retail, Nov 14; Inmobiliaria Sur, Nov 21) and one waiting for the court's reply.Cases · rows 4, 9, 12Agenda · Nov

Prepare the hearing brief for the Andina case.

get_document(Andina file) · search_knowledge_base(evidence) · create_document

Done. I created “Hearing brief · Andina Retail” with the facts, the claims and the 6 exhibits cited by block. Only people with a role in Legal can see it.

✓ document created · Legal / Hearings · trail #c21e

Measured with our own evaluation harness — no customers to quote yet, so we show numbers.

See the method

100%search recall on the evaluation set (was 87.5%)

60%of tasks completed by the agent on a local 4B model (was 22%)

100%of answers in the asker's language on a local model (was 32%)

~3 msper query for hybrid keyword + semantic fusion, no GPU

the differentiator

A process no longer needs a system. It needs a knowledge base and an agent.

Today every internal process (leave, certificates, appointments, orders) means building a system: data model, screens, permissions, integrations and maintenance. Or it lives in email and spreadsheets. Doctiling changes the starting point: the information is already organised in knowledge bases with their own access hierarchy. The tenant is the agentic semantic layer that queries and manages them. A small agent, connected to the right base, is the process.

A system of screens

per process

data model · forms · states · approvals · integrations

months of work · permanent maintenance · one more system to learn

An agent over the base

doctiling

I want 5 days of leave starting November 18.

search_knowledge_base(Policy) · aggregate_rows(Leave) · append_table_rows

You have 14 days available and it does not overlap the accounting close. I recorded the request for Nov 18–22 as pending and notified your manager.Policy · §3Leave · row 12

✓ row added · People / Leave · trail #a41f

the base already exists, with its permissions · the agent connects over MCP with a scoped token

Hours, not months. No new data model, no new screen, no hand-written query.

The tenant stores no data: it is the semantic layer that understands, applies permissions and acts. The hierarchy starts in each knowledge base: its own roles (owner, admin, collaborator, reader) and, per document, who sees and who writes, plus the classification (public, internal, restricted, private) the server enforces. Adding a process is adding a base; the agent only connects to it.

benchmarks

Measured numbers, not promises.

Every quality change ships with a before-and-after measurement. These are the two that shape what you see most: finding the right passage, and the agent finishing the task.

Search: finding the right passage

12 questions · known answers

Weighted keyword + semantic fusion and a threshold calibrated for the production embedder, against the previous version. Same set, same embedder.

Recall

before0.875

now1.000

nDCG@10 · result order

before0.878

now0.939

MRR · first useful result

before0.917

now0.944

MAP · mean precision

before0.861

now0.906

Agent: finishing the task on a small model

10 runs per task · local 4B model

The knowledge-base agent running on a 4B open model on a local machine, with the product's 28 MCP tools. Before and after tightening the tool contract and the reply language.

Tasks completed

before22 %

now60 %

Replies in the asker's language

before32 %

now100 %

~3 msper query for hybrid fusion, CPU only (50 + 50 candidates).

Measured with our internal evaluation harness. The sets are small and our own: we publish them as evidence of improvement, not as a ranking against third parties. We re-measure with every search or agent change.

security & governance

AI your security team can sign off on.

Governance enforced by the product, not a policy PDF: you decide who gets in, what leaves and what the agent may do — and every action leaves evidence.

Your directory, your access rules

Sign-in through your corporate identity provider and its MFA, roles per workspace and per knowledge base, session limits, and signed access reviews ready for your audit.

Confidential stays inside

Classify knowledge bases and documents. Restricted content is never published, exported, served over MCP or sent to an external AI: the server blocks it, not good intentions.

An audit trail nobody can edit

Every sign-in, read, export, agent query and MCP call is recorded for at least a year, and the database refuses to alter it. Export it or forward it to your SIEM.

Backed up, recoverable, inside your boundary

Encrypted backups that restore and verify themselves every week. And when your data must stay home, self-host the very same product.

NIST CSF 2.0
The product's cybersecurity profile, function by function, published and self-assessed.
Self-assessed
ISO/IEC 27001:2022 · Annex A
Control-by-control mapping to Annex A; no certification audit yet.
Mapping publishedNot certified
Personal data (Ley 1581 · GDPR)
Personal-data processing under Law 1581 and the GDPR, with exercisable rights.
In force
Data Processing Agreement
Data processing agreement available to every tenant, sub-processors listed.
In force

Open the trust center: every claim with its evidence →

platform

Built to outlast this season's model.

each base routes toCloud providerLocal open modelA different one per base

Any model

Switch providers without migrating anything, assign a model per base or run open models on your own machine. When the frontier moves, you move with it.

Text, tables and graphs

Three living document types that experts and agents edit over the same base, with 17 one-click analysis lenses.

use caseContract review · legal teamYour teamDoctiling team

Tailored support

On Platinum, a development team works with your experts to adapt the core to your workflow: prompts, tools and integrations.

interfaces

Use your knowledge where you already work.

Real capture of the studio: the Nortera Logística base with the Q3 quarterly report open and its summary citing each block in the mirror view.

The studio: your documents, tables and graphs on the left; the document in the middle; and the agent's analyses on the right, each citing the exact block.

for your field

Your field already lives in documents. Doctiling makes them answer.

Not a generic chat that guesses: every knowledge base is grounded in your own sources and answers with the exact citation. We set up your space with knowledge bases, skills and starter documents tuned to your specialty.

Legal teams

Review contracts and draft matter briefs without losing the exact clause. Every conclusion cites the document and page.

your edge

Answers you can defend to a client or a judge. On Gold your matters live in a dedicated database — never mixed with another tenant.

Healthcare

Bring patient history, guidelines and literature into one knowledge base. Suggest differentials and summarize follow-ups, all cited.

your edge

Never invents a dose or a finding: if it isn't in your sources, it says so. End-to-end clinical traceability.

Finance

Ratio analysis, filing summaries and portfolio tracking over your own filings and statements.

your edge

Every figure points to the statement it came from — audit-ready, with no fabricated numbers.

Accounting

Classify transactions, reconcile and query the general ledger by asking in plain language.

your edge

Full traceability: every answer links back to the original entry. Close stops being a hunt for receipts.

Research & grad school

Turn a stack of papers into a literature review: synthesize across sources, surface contradictions and draft.

your edge

Verifiable citations, never invented references. A bilingual studio (EN · ES) so you can publish in either language.

Consulting

Turn data rooms and interviews into grounded deliverables. Mirror view stacks summaries, entities and correlations beside the document.

your edge

Save each finding as a variant and expose it to your team chat over MCP — the client sees the trail, not a black box.

See solutions by field in detail

multitenancy

Pick the isolation level that matches your risk.

Doctiling runs as a multitenant studio. Free access is sponsored and invitation-only — request a demo and we activate your seat. Paid tiers move you to your own database; Platinum adds dedicated infra and a development team that can tailor the core to your workflow.

tier 0

Free · Sponsored seat

Sponsored free access, by invitation. Request a demo and we activate your seat in the shared studio with a sponsored AI budget — no credit card. Your data stays logically scoped per user.

By invitation — request a demo to activate your seat.
Sponsored AI budget · 25 AI calls/day · 100/week.
All Free-tier limits enforced server-side.

Request access

Gold · Private studio

On request

Your own dedicated database — isolated from every other tenant. Same hosted runtime, but no neighbour shares your rows or embeddings. White-label branding is included.

Dedicated database, schema migrations on your timeline.
Custom domain + white-label branding.
Higher quotas, priority support, dedicated onboarding.

Request a demo

tier 2

Platinum · Sovereign studio

On request

Dedicated database AND dedicated infrastructure. Plus a development team that works alongside your stakeholders — humans and AI agents — to modify the Doctiling core: prompts, tools, integrations, UI surfaces. Built for organizations that need the engine, not just the wrapper.

Dedicated database + dedicated runtime hosts.
Full white-label, customizable core (prompts, tools, UI).
Dedicated dev + AI-agent squad attached to your tenant.

Talk to us

request access

Request your free access.

Tell us who you are and what you want to ground in cited documents. We'll review your request and email you an invitation — your sponsored free seat, no credit card. Teams can ask about Gold and Platinum here too.

Sponsored free access · by invitation · no credit card

questions

Everything you'd ask before signing in.

ready when you are

Your team already knows. Put that knowledge to work.

Request your free access with your email — we review it and send the invitation shortly. No credit card. Already invited? Sign in and your first knowledge base is one click away.

Request free access I already have an account

Support · support@doctiling.app