Conversation ended
Lina brings OpenAI models to your website: visitors can find information, compare offers and carry out authorised actions in your systems through conversation. Five components work simultaneously in one session: voice and Liveboard connect Lina to the visitor, tools connect her to the website’s systems, and intelligence and context sit in between.
Lina holds a natural conversation in real time.
Lina presents information on a surface next to the conversation. The visitor makes selections there by tapping, and Lina picks up the selection in the conversation.
Lina consists of two models. The speech model leads the conversation; the backend work model handles multi-step tasks.
Speech, displayed scenes, selections and tool results work together. You decide which content is retained beyond the individual conversation.
Connect Lina to your shop, CRM, calendar or custom software. Our tool gateway executes the functions you authorise and brings their results back into the conversation.
What a conversation with Lina looks like on your website. Three capabilities in three examples – applicable to any website where people want advice.
A watch boutique: Lina asks about occasion, style and budget, compares suitable models and reserves a fitting in the boutique.
A car dealership: Lina puts the specification together in conversation, recalculates the lease on every change and books the test drive.
A new-build project: Lina shows suitable apartments, roughly calculates the financing and arranges a viewing at the show home.
The retailers and the housing project are examples. The watches and the car are real products with prices from the manufacturers or the ADAC catalogue (October 2026); the lease is an example calculation. Which data and actions Lina uses on your website is defined by your tools.
Erfindergeist · Architecture
Erfindergeist brings real-time conversations, generative user interfaces and AI agents that execute actions together in a unified, embeddable runtime for digital customer interactions.
Erfindergeist provides hosting, the conversation runtime, model access, monitoring and central updates. You configure your Lina in a versioned profile and connect the authorised functions of your systems as tools.
Run by ErfindergeistControlled by you
The model key stays with us. Your developer key lives in the development client, the separate backend key in your backend. The visitor’s browser only receives the short-lived session token.
Set up Lina with Codex, Claude Code or your development team directly in your existing website project. A documented API, management MCP and website starter provide the path. We run the Lina server and provide model access.
Create a developer key in the customer area and set up the API or MCP in your coding agent. It can use the existing project context.
Define profile, voice, startup context and your own tools. The coding agent can create the configuration through the same documented contract.
Adapt the starter for the SDK and session endpoint to your website. Visitor access and authorised context are confirmed in your own backend with a separate key.
Confirm the domain by DNS, check readiness and publish the app version explicitly. A conversation only starts when the visitor starts it.
[mcp_servers.lina_live]
url = "https://app.erfindergeist.ai/v1/live/mcp"
bearer_token_env_var = "LINA_LIVE_API_KEY"
startup_timeout_sec = 20
tool_timeout_sec = 75
{
"mcpServers": {
"lina_live": {
"type": "http",
"url": "https://app.erfindergeist.ai/v1/live/mcp",
"headers": {
"Authorization": "Bearer ${LINA_LIVE_API_KEY}"
},
"timeout": 75000
}
}
}
curl -X POST https://app.erfindergeist.ai/v1/live/ops/applications.list \
-H "Authorization: Bearer $LINA_LIVE_API_KEY" \
-H "content-type: application/json" -d '{}'
Every operation as an HTTPS call with a bearer key; inputs and outputs are in the OpenAPI reference.
Open the Quickstart ↗In the customer area you follow running sessions and open authorised conversation histories. You decide which content is stored and for how long; access follows roles and permissions.
Protection along the whole conversation: every station checks on its own, none relies on the one before.
Run by ErfindergeistControlled by you
€299
per month excluding applicable taxes
500 minutes included
Get LinaSetup with us or individual requirementsAdditional minutes: €0.29, prepaid and billed by the second. Included minutes expire at the end of the period; purchased extra credit carries over within the active subscription.
Cancel monthly, effective at the end of the paid period. No setup fee when you set it up yourself, no fee per connected tool. Optional setup assistance is agreed separately. You cover your own external services and integrations.
For setup with us and for requirements beyond the standard plan.
Scope and terms are agreed in advance. Enquiries are without obligation.
Email[email protected]We will get in touch personally. Nothing has been booked.
No. Model access is included.
Shops, CRMs, calendars, data sources or custom applications: Lina calls suitable HTTPS or remote MCP tools through our gateway. Existing APIs and integration services can provide the connection. Available functions depend on your software’s interfaces and the access you authorise.
Yes. Set instructions for the speech and work models, voice, startup context and supported model parameters. Your own tools use your descriptions, schemas and permissions. We maintain the shared Liveboard components and platform tools.
We version the shared platform core, Liveboard tools and matching SDK together. With the Hosted bootstrap, your website loads the matching SDK version on the next page load. Your app profile remains under your control. Manually pinned SDK versions need a coordinated update.
Yes. One app can combine, for example, advice, appointments and service requests. Separate apps make sense for different audiences, instructions or access boundaries.
A current browser and microphone permission. No installation.