No vendor lock-in

Our thesis has two halves. On the model: “We guarantee the capacity, not the provider.” On your data: it’s yours, always, and you can take it with you whenever you want. Lock-in (getting trapped in a platform because leaving costs too much) is a design decision, and at Shara we’ve decided the opposite. This page explains, with no fine print, why you don’t stay captive with us.

The two faces of lock-in

When someone fears getting trapped in a tool, they actually fear two different things. Worth separating them because they have different solutions:

• Model-provider captivity. That the platform ties you to whichever AI engine is in use, and that if that provider raises the price, degrades quality or leaves the EU, it becomes your problem. We solve it with model aliasing.
• Data captivity. That your information ends up locked inside the platform in a proprietary format, so leaving means starting from scratch. We solve it with open export and real portability.

Neither is a favour we’re doing you: they’re contract clauses and verifiable technical decisions. If one day we stop convincing you, we want the exit door to be open and well signposted.

How aliasing works

Your company always talks to four public models. The real provider behind each alias is irrelevant to your day-to-day and changes whenever a better one appears, without your team touching anything.

Alias	What for
Prelude	High volume, low cost: replying to email, classifying a lead, validating a field.
Sonata	Medium reasoning: proposals, reviewing a standard contract, planning the week.
Symphony	Advanced reasoning, available from Max (strategy, sensitive legal). It is Amadeus’s model on Max and above.
Concerto	Concerto Local’s local engine: the model, the agents and the data, 100% on your infrastructure (sign-in goes through the cloud).

The real provider behind each alias is confidential. Max and Concerto Local clients can request it under an NDA, through support@aiginer.com.

What it means for you

• No mandatory migrations: your prompt works today and in a year.
• No negotiating with a third party: you renegotiate only with AIGiner, not with each provider.
• No regulatory surprises: if a provider changes its data policy or leaves the EU, we migrate without your team ever knowing.
• Consistent capacity across plans and over time.

Guarantees we sign

• Capacity continuity: replacement with a model of equal or higher performance, at no cost.
• No silent degradation: 30 days’ notice if the quality behind an alias were to drop.
• Internal traceability: every call logs which alias was used and what cost it generated.

Your data is yours

Let’s start with what matters: ownership of your data doesn’t change by using Shara. You remain the data controller; we act as data processor on your behalf, under GDPR and the data processing agreement (DPA) you sign when activating the account. We don’t sell your data, don’t use it to train models and don’t mix it with other clients’ data: multi-tenant isolation keeps it separated at all times.

We consider yours and exportable everything you’ve put or generated inside the platform:

• Your corporate identity (IDENTITY.md): tone, values, rules and its patch history.
• The conversations with Amadeus and with each departmental agent.
• The drafts, proposals and deliverables produced by the agents.
• The full audit log (who asked for what, which agent resolved it, before and after every action).
• The configuration: active agents, departments, roles, team members and schedules.
• The historical STU consumption and the billing detail.
• The list of linked connectors (references and metadata, never your passwords, which we never see).

Inference and the database live in the EU. Export isn’t an exceptional procedure you have to justify: you do it yourself from the panel.

Export whenever you want

An owner or administrator can export all of the company’s data from the panel, under Settings → Organization → Export all data. You download a .zip with one file per table in JSON Lines format (.ndjson, one record per line) and a manifest.json that lists, for each file, how many rows it holds and its sha256 hash so you can check it hasn’t been altered. It can be done once every 24 hours per company: if the download is cut or fails halfway, it still counts and cannot be repeated until 24 hours have passed. When it happens, all owners and administrators are notified, including the person who requested it, and it is recorded in the audit log.

Only what you can see in the app today comes out. The export does not include secrets (connector tokens, API keys, credentials or webhook secrets: for each connector only the provider, status and dates are exported), nor anything the app does not show you: private conversations and personal memory of other team members, other people’s private flows, what the agent did on other people’s devices, and other departments’ channels and memory; your own are included. If you need another format, or something the button doesn’t cover, write to privacy@aiginer.com and we’ll handle it within the timeframes set by GDPR.

Open formats, nothing proprietary

There’s no file encrypted with a key only we have, and no binary format that only Shara can read. Everything comes out in formats any tool, script or person can open. In the company export, every table comes out as NDJSON with a manifest.json; CSV is available on request.

Data	Format	What opens it
Corporate identity	Markdown (.md)	Any text editor.
Conversations and audit	NDJSON (.ndjson)	One record per line; any language or jq.
Drafts and deliverables	Markdown / original attachments	Text editor, office software.
Configuration and connectors	NDJSON / JSON	Any standard parser.
Consumption and billing	CSV (on request)	Excel, Google Sheets, pandas.

We chose NDJSON, JSON, Markdown and CSV on purpose: they’re open, plain-text and widely supported formats. Nothing you take with you depends on Shara continuing to exist to be able to read it.

Real portability

Portability isn’t just being able to download a file: it’s that what you download is useful somewhere else. That’s why Shara is built on open standards at every layer that connects you to the outside:

• OAuth connectors, not resale. Your integrations (email, calendar, CRM, storage) are yours: you authorize Shara via OAuth and revoke that access at the provider itself whenever you want. There’s nothing to “get back” because it never stopped being yours.
• Tools via MCP. The tool bridge uses Model Context Protocol, an open standard. What you expose to the agents doesn’t stay locked in a Shara-proprietary format.
• Identity in Markdown. Your IDENTITY.md is a human-readable text file; you can copy it, version it and edit it outside Shara.
• Plain REST API. The REST API is standard, with JSON and normal headers. An API key lets you invoke agents and use the MCP bridge; the rest of the panel needs a signed-in session. No mandatory SDK or closed protocol.

What you take with you if you leave

Let’s be concrete. The day you decide to leave, this is exactly what walks out the door with you:

• Your whole corporate identity and its change history, ready to reuse in any system.
• The full conversation history with Amadeus and the agents.
• All the drafts and documents the agents produced for you.
• The full audit log, which is your defensible proof for a client or an inspection.
• The consumption and billing in CSV for your accounting.
• The configuration of your team, agents and departments, in case you rebuild the flow elsewhere.

And a hygiene promise: when you close the account, after the retention period agreed in the DPA, we delete your data from our systems. We don’t keep a copy “just in case”.

And, being honest: what you don’t take with you

Honesty is part of not tying you down. There are things that, by their own nature, aren’t part of the export, and we’d rather say it clearly before you discover it yourself:

• The AI model weights. On cloud plans, the engine is served by an external provider under its own license; it isn’t ours to give you, just as it isn’t any other platform’s. What is yours is everything that engine produced while working for you. If you need the model inside your own house, Concerto Local exists (below).
• The orchestration itself. Amadeus and the agents’ behavior is a Shara product. You take its results and your configuration, not the code that coordinates them.
• Your connectors’ live tokens. They aren’t exported because they aren’t data: they’re third-party credentials you manage and revoke at the source provider (Google, Microsoft, your CRM).

Concerto Local: total independence

If your threshold for peace of mind is “nothing depends on someone else’s servers”, the answer is Concerto Local: Shara deployed 100% on your own infrastructure, on-premise, with the local Concerto engine. It’s the strongest way to have no lock-in, because it removes the dependency on our servers at the root.

• Everything at home. Data, model and execution live on your machines. Nothing goes out to the network except what you decide to connect.
• Reasoning stays at home. Concerto Local doesn’t call our servers to reason. It checks its license every 24 hours; if it can’t connect, it keeps working for 30 days and then degrades in stages (see Concerto Local).
• Local inference. Concerto runs locally, so the model itself does not need the internet. Your connectors (email, CRM…) still do.
• Same product. The same agents, the same corporate identity, the same auditing, the same API. What changes is where it runs, not how it’s used.

Concerto Local is built for those with strict data sovereignty or compliance requirements.

Quick comparison

	Shara	Traditional platforms
Model you see	Prelude, Sonata, Symphony, Concerto	The current model’s commercial name
Switching to a better model	Automatic, nothing to touch	Mandatory prompt migration
Price negotiation	Only with AIGiner	With each provider
Risk if the provider raises prices or lowers quality	Zero: we absorb it	You absorb it
Your data	Yours, exportable whenever you want	Often in a proprietary format
Export format	NDJSON, JSON, Markdown, CSV	Variable or non-existent
On-premise option	Yes (Concerto Local)	Rare
GDPR / EU AI Act compliance	Guaranteed by contract	Depends on the provider

Frequently asked questions

• How do I know the model is good? Every agent has an eval suite that runs before any change is promoted; if measured quality drops, it isn’t applied.
• Can I know the provider? You choose the alias, not the provider. Max and Concerto Local clients can request it under an NDA through support@aiginer.com.
• Does the export cost money or require permission? No. It’s a GDPR right: you do it yourself from the panel, with no permission or justification needed (if you need something the button doesn’t cover, write to privacy@aiginer.com).
• Is the exported data useful outside Shara? Yes. It comes out in open formats (NDJSON, JSON, Markdown, CSV) that any tool can open, with no dependence on us.
• What happens to my data if I close the account? It’s deleted after the retention period agreed in the DPA. We don’t keep silent copies.
• And if your company disappears? Concerto Local runs on-premise with Concerto with no dependence on our servers; in the cloud, you can export all of your data from the panel at any time.
• Why classical-music names? They’re neutral, memorable and reflect the hierarchy: Prelude, Sonata, Symphony, Concerto.