Evidence before intelligence

Your company trusted memory, ready for safe AI work.

Avrelia turns your documents, decisions and conversations into one working memory where AI agents propose, your team approves, everything is recorded. No heavy integrations or long onboarding.

A claymation scene of the Avrelia operating layer: an open book with a brain and a sealed checklist, a server tied to a cloud, a small vault, a mailbox, a calendar and a notifications panel.
Event stream · live14:22:41Z
MSA · MeridianPayment · risk
Invoice · Q2-4411Net-60 · mismatch §3.1
Truth-report3 obligations slipped
Email · bidsAction proposed
Isolated data
Any model
Any agent
§ 01 — Why this matters

Every company has a memory problem.

In the age of AI, the truth about what really happens is the ultimate currency. What you agreed, with whom and why — that knowledge sits in heads, inboxes, chats and drives. Nobody holds the whole picture, and you have to decide anyway.

A claymation ring of fragmented documents, charts and a checkmark — work scattered across systems.
01
Work happens
in fragments
Contracts in storage. The decision that changed them lives in a chat thread. Letters in inboxes, unread. Each system knows only its slice.
A claymation chat bubble being erased by a red eraser, with a pin marker beside it — assistants forgetting context.
02
Assistants forget
between answers
Confident answers come back with no source. Same question tomorrow, same blind spots, the same context tax paid again.
A claymation wind-up car carrying chess-piece cargo with its key on the floor — autonomous agents running without audit.
03
Agents act
without audit
An agent that applies a change the moment it decides leaves no before, no after, no approver. By the time you notice the loss, there's no way to replay what happened.
§ 02 — The core idea

Your company is not a prompt.

It is a living record of documents, decisions and obligations — lived through email, chat and a dozen other places. Two ways to use it: Insight surfaces what's actually happening; Operate drafts the documents, recommendations and changes your team can review, approve and apply.

Lane A · Insight
Ask. Verify. Get the truth.
Read-only · evidence-cited · always on
A fast assistant over your shared memory, plus scheduled truth-reports — daily, weekly, monthly — that show what actually happened on your projects, not what people said in the standup.
Ask the company anything
Daily, weekly, monthly truth-reports
Citations under factual answers
Lane B · Operate
Drafts and actions.
Generates work · approval-aware · auditable
Drafted documents, recommendations and early interventions — proposed before a tone shift, a slipped obligation or a quiet money leak gets expensive. Automated checks run on every change; the material ones need a human approver.
Generates documents, not just answers
Structured changes under policy and approval
Undoable changes
§ 03 — Architecture

Six layers, cleanly bounded.

The AI market moves too fast to bet your architecture on one model or one kind of agent. So each layer stays separate — when a better model ships, a different agent fits, or your data must move in-house, you change one piece on the same data, never the stack.

01
Settings & routing
Company account, prompt and task templates, LLM routing.
02
Ingestion & structuring
Collected from your systems and sources, with formats unified.
03
Company memory
Document indexes, relations, primary-source citations, summaries.
04
Retrieval & evidence
Cross-object search, excerpts from originals, freshness controls.
05
Proposal & approval
Risk analysis, drafted recommendations, document drafts.
06
Execution environment
A secure runtime for AI agents on structured data, with external integrations.
§ 04 — Lifecycle

Ingest, structure, explore,
propose, apply.

The lifecycle turns scattered incoming materials into a reliable working library: first the system separates business context from noise, then extracts text, links entities and prepares the evidence agents need to complete tasks safely.

01
Ingest
Everything you receive lands here — emails, chats, contracts, uploads. Files stay quarantined until reviewed.
02
Structure
Text becomes facts, indexes and summaries — regenerable, inspectable.
03
Explore
One governed search across documents, decisions, conversations and memory pages — every result with its citation.
04
Propose
Agents draft documents and changes. Nothing reaches your business until you say so.
05
Apply
You approve. The system applies the change and lands it in the ledger.
Action proposal · awaiting review
Change payment term on Meridian MSA: 60 days30 days
evidence · 3 citations
policy · legal.msa.v3
proposer · agent/contract.review
checks · ok · automated
provenance · append-only
14:22:11proposal.createdagent/contract.review
14:22:18evidence.attached4 citations · doc-4421 §3.1
14:22:19checks.okautomated · policy.msa.v3
14:22:41approval.pendingS.Park · legal-ops
§ 05 — Governed execution

Agents propose.
Humans approve.
Everything is recorded.

Avrelia turns AI work into managed business processes. An agent can flag risks in a contract, draft the edits, attach the rationale and send the package to the right person for approval. What used to take days happens in hours — without losing control over the key decisions.

§ 06 — Deployment & sovereignty

Your data stays
exactly where you put it.

Multi-tenant SaaS, dedicated cloud, or self-hosted — including completely isolated inside your network with no outgoing connections. Built for industries where the data and the chain of decisions cannot leave your perimeter.

Claymation deployment & security illustration
Data isolation
Databases and file storage — on the servers of your choice.
Model-agnostic
OpenAI, Anthropic or open-source models.
Pluggable agent
Codex, Opencode, OpenClaw, Hermes — or pick another.
Secure execution
Each task runs in its own environment — data and resources never overlap.
Custom solutions
Create new tasks and refine existing ones for your industry and processes.
Full autonomy
Deployment option that needs no external internet access on request.
§ 07 — Where it lands

Where losing the truth costs the most.

Preventing a problem costs many times less than proving you were right after the fact. A forgotten clause, an unsigned acceptance, a missed deadline or a thin paper trail turn into fines, disputes and lost customers — even risk to the whole business. It hits hardest in fields built on official correspondence, written agreements and formal procedure: government contracts, large holdings, regulated markets. Avrelia surfaces these risks earlier and preserves the evidentiary trail.

01
Legal defense
Contracts, amendments, redlines, clause libraries — with the evidence pack ready when a dispute lands.
letters, claims, suits
02
Finance & controllership
Invoices reconciled against contracts and delivery — fewer wrong payments, cleaner audits.
AP, AR, audit
03
Operations management
What policy says should happen, what actually happened, and when the alarm should fire.
policies, reports, KPIs
04
Sales & development
Customer, market and conversation history — so you can tailor outreach and decide the next step.
CRM, markets, clients
05
Procurement & bids
Counterparty memory, renewal calendars, pricing history — for your own buying and for bidding.
RFPs, vendors, bids
06
People operations
Policy retrieval, case precedent, structured proposal workflows — and quiet nudges when work goes unlogged.
ER, L&D
§ 08 — Pricing · Annual contracts

Priced for the work, not the seat count.

Insight
$2,500+/month
Shared company memory plus a fast assistant. Clear answers from your own data — on demand, on schedule, with citations.
Shared memory + Insight lane
Daily, weekly, monthly truth-reports
Annual contract · no separate setup fee
Book a pilot
Recommended
Operate
$3,500+/month
Both lanes, end to end. Agents propose, your team approves what matters, the system applies — full audit, replayable.
Insight + governed changes
Document & report generation
Annual contract · no separate setup fee
Book a pilot →
Sovereign
Quote
A custom build: multiple legal entities in one workspace, dedicated infrastructure, and feature work tuned to your business processes.
Private cloud, on-prem or fully isolated
Custom tasks and AI-agent capabilities
Upfront fee for the custom setup
Book a pilot
Pilot

Month-long pilot on one real project.
If it doesn't deliver, you don't pay.

Scale

Scales with project count and data volume.
LLM usage is bundled; heavy usage is billed extra.

§ 09 — Next

Bring us one
messy workflow.

Dmitriy Surkov, founder of Avrelia
Dmitriy Surkov
Founder, Avrelia
LinkedIn →
I built Avrelia after my company nearly went under on a public-sector contract — the counterparty disengaged, blamed us, then terminated unilaterally. We saved it by proving our integrity and reconstructing the truth out of inboxes, chats and people's heads. Avrelia would have flagged the drift early and built the evidence pack in hours.
Let's talk — and if we're not the right fit, I'll tell you in the first 30 minutes. If we are, a month of weekly truth-reports on one real project — ready for your board.
Request a pilot