Oriveo brings AI models together in a local client

What this is about

Oriveo Community Edition is an open-source multi-model client for iOS, Android and the web. Users provide their own API keys or connect local model servers such as Ollama, LM Studio, llama.cpp and vLLM. It is aimed at people who want to use several AI services in one interface without necessarily storing their conversations with an additional chat provider.

The project appeared in September 2026 and is licensed under AGPL 3.0. Version 1.1.0, released on September 23, introduced the first directly installable Android APK. The iOS and web applications in this Community Edition still have to be built from source. Proprietary Oriveo apps with accounts, cloud sync and managed models are offered separately. That distinction matters because their privacy properties and feature sets are not identical.

What Oriveo actually does

Oriveo provides one chat interface for 15 officially supported providers and compatible endpoints. According to the project, those include OpenAI, Anthropic, Google Gemini, Mistral, Groq, DeepSeek, OpenRouter and xAI. Users can configure their own keys, models and parameters for each provider. A response can be handed to a second model for cross-checking.

Conversations, notes, folders, skills and attachments stay on the device by default in the Community Edition. The web version can be self-hosted. It still needs its own server component for many providers because browsers cannot call their APIs directly. Oriveo can also connect remote MCP servers. Each tool can be configured to ask before execution, run automatically or remain disabled.

Other features include search, attachments, cost tracking based on reported usage, encrypted backup of optionally included API keys and connections to local models on the user's network. The software is only the client. It does not run large language models on the phone itself.

Why it matters

People who test several model providers often spread conversations and settings across multiple apps. Oriveo brings those accounts together while promising no additional markup on provider prices. For teams and demanding individual users, it is particularly useful that the same conversation can be reviewed by different models and that local endpoints appear alongside cloud services.

The open source code also makes the client inspectable. AGPL permits use, modification and self-hosting, but requires corresponding source code to be made available when modified software is offered over a network. Companies should therefore review the license before deploying a customized version. Security remains practical work as well: a multi-model client handles valuable API keys and can trigger actions through MCP tools. Client safeguards do not replace provider spending limits or carefully restricted tool permissions.

In plain language

Oriveo is like a power strip for AI services. Instead of finding a different wall socket for every device, several connections meet in one place. The power strip does not generate the electricity: costs, availability and rules still come from each provider. Anyone connecting a powerful appliance still needs circuit breakers and must know where the cable leads.

A practical example

A developer uses three services each month: one model for code, one for long documents and a local Ollama server for confidential drafts. She creates three provider profiles in Oriveo. For a ten-page project plan, she starts with the local model, sends only a sanitized summary to a cloud provider and asks a second model to review its answer.

She enables an MCP tool for internal knowledge search but requires confirmation before every run. After 20 requests, she sees the usage reported by each service in one interface. This makes comparisons easier but does not replace the provider's invoice. Before changing devices, she exports her data and protects included keys with her own password.

Scope and limits

First, locally stored does not mean entirely locally processed. Cloud models receive the prompts sent to them. In the web application and for certain features, requests may also pass through Oriveo servers. The privacy policy explicitly documents these paths.

Second, cloud sync in the proprietary apps is not end-to-end encrypted storage. According to the privacy policy, synchronized messages and metadata are stored in readable form in Firebase. Users seeking maximum control should self-host the Community Edition, leave sync disabled and choose local models.

Third, broad integration increases the attack surface. Misconfigured compatible endpoints can see keys, MCP tools can perform real actions and model output remains fallible. Organizations should restrict provider credentials, set spending limits, grant tools minimal permissions and review the AGPL implications.

SEO & GEO keywords

Oriveo, Oriveo Community Edition, multi-model client, BYOK, local AI, Ollama, MCP, self-hosted AI, AI chat client, AGPL 3.0, API keys, privacy