A local AI gateway for your apps.

Not a chatbot. A local AI gateway.

Froggy Context Hub is a local AI gateway that gives your apps access to private knowledge, MCP tools, prompt profiles, and grounded context — without rebuilding RAG for every client. Instead of rebuilding retrieval, prompt profiles, MCP wiring, and context injection in every client, point your app at Froggy. One local pipeline handles private knowledge and grounded context for all of them.

Your app → Froggy Context Hub → Your LLM
with the right context already injected

Intercept → Retrieve → Inject → Forward

Point your app at Froggy. Get context automatically. Froggy Context Hub is a serious local context gateway — packaged as a desktop app with visible controls, not a chat box with hidden plumbing.

Why "Context Hub"?

Because the product is bigger than RAG alone. Froggy is a local AI gateway for private knowledge, MCP tools, prompt profiles, and grounded context — plus namespaces, passthrough APIs, optional web search, external MCP coordination, and policy controls.

Developer-grade context infrastructure. Desktop-app simple.

Most local AI tools make you choose between power and usability. Froggy Context Hub gives you both: a serious local gateway for RAG, passthrough APIs, MCP search, tool coordination, retrieval tuning, metadata filtering, and diagnostics — packaged as a desktop app with visible controls and test surfaces.

Froggy does not hide the important parts. You can see what is indexed, test retrieval, inspect server status, tune chunking, configure passthrough, enable MCP, manage external tools, and watch usage stats from one UI.

General settings

A real installed desktop app with tray behavior, startup options, background indexing controls, and update checks.

Why it matters: Froggy runs like normal local software — not a fragile script you babysit.

Server stats

Observe requests, latency, chunks returned, web searches, corpus searches, passthrough chats, MCP calls, and namespace usage.

Why it matters: Debug the local context layer with numbers — not guesswork.

Most RAG setups push complexity into every app. Froggy centralizes it.

Intercept → Retrieve → Inject → Forward

Point your app at Froggy. Get context automatically. Your app → Froggy Context Hub → LLM, with context already injected.

1. Intercept the request

   Froggy Context Hub accepts OpenAI-compatible or Ollama-compatible requests from existing tools and clients.

2. Search the active namespace

   The local RAG engine retrieves ranked chunks from the selected corpus using the configured retrieval behavior.

3. Add the right context

   Froggy injects structured context into the prompt so the model can answer from your knowledge bases.

4. Optionally coordinate tools

   When configured, Froggy can blend web search or MCP tool results into the same flow.

5. Forward to your model

   Froggy sends the enriched request to the upstream LLM you configured, local or remote.

6. Return the answer

   Your app receives the response through the API shape it already expected.

See the setup guide for installation, namespace configuration, and passthrough listener setup.

Search test

Validate retrieval quality inside the app before connecting external clients.

Why it matters: Tune chunking and retrieval with immediate feedback — not after wiring every client.

Passthrough test

Run an end-to-end RAG plus upstream LLM test on the same path inbound passthrough requests use.

Why it matters: What you test in the UI is what your OpenAI- and Ollama-compatible clients get.

Use your existing clients. Change the base URL.

Froggy Context Hub speaks the API shapes developers already use. OpenAI-style clients can call the OpenAI-compatible listener. Ollama-style tools can call the Ollama-compatible listener. MCP clients can call rag_search. The context pipeline stays consistent across all of them.

Instead of sending requests directly to your model, point the base URL at Froggy. Your client keeps its familiar request shape; Froggy adds the missing context layer.

• Existing OpenAI SDK-style apps
• Local Ollama workflows
• Scripts and internal tools
• Cursor and Claude Desktop through MCP
• Developer tools that need rag_search
• Apps that should gain context and RAG without embedding retrieval logic directly

LLM passthrough settings

Configure Froggy to sit between OpenAI-compatible or Ollama-compatible clients and your upstream model.

Why it matters: Change the base URL in your client; Froggy owns the context injection layer.

Local-first by design.

Froggy Context Hub runs as a local desktop app with its own UI, API surfaces, vector store, ingestion controls, passthrough listeners, MCP configuration, and server stats. Your documents are indexed locally unless you explicitly configure remote model providers, web search, or external tools.

• Local desktop bundle with Windows installer and auto-updates
• Local vector store and namespaces
• Configurable ports and upstream model providers
• Local API tests, server status, and stats
• No separate server farm required for personal or small-team use

Folder ingestion

Point Froggy at local folders and source trees — no custom ingestion scripts required.

Why it matters: Real corpora land in the hub through a visible UI, not hidden pipeline code.

Vector store

See indexed documents, chunk counts, metadata, and search readiness instead of a black-box index.

Why it matters: You can inspect what is actually in the corpus before trusting retrieval.

See the local context hub in action

Every screen below supports a specific claim: visible ingestion, inspectable indexes, built-in test tools, passthrough configuration, MCP exposure, retrieval tuning, optional web search, and local observability — not decorative UI chrome.

General settings

A real installed desktop app with tray behavior, startup options, background indexing controls, and update checks.

Why it matters: Froggy runs like normal local software — not a fragile script you babysit.

Folder ingestion

Point Froggy at local folders and source trees — no custom ingestion scripts required.

Why it matters: Real corpora land in the hub through a visible UI, not hidden pipeline code.

Vector store

See indexed documents, chunk counts, metadata, and search readiness instead of a black-box index.

Why it matters: You can inspect what is actually in the corpus before trusting retrieval.

Search test

Validate retrieval quality inside the app before connecting external clients.

Why it matters: Tune chunking and retrieval with immediate feedback — not after wiring every client.

Passthrough test

Run an end-to-end RAG plus upstream LLM test on the same path inbound passthrough requests use.

Why it matters: What you test in the UI is what your OpenAI- and Ollama-compatible clients get.

LLM passthrough settings

Configure Froggy to sit between OpenAI-compatible or Ollama-compatible clients and your upstream model.

Why it matters: Change the base URL in your client; Froggy owns the context injection layer.

Inbound MCP server

Expose local corpora to MCP clients such as Cursor and Claude Desktop through rag_search.

Why it matters: IDE and agent workflows get local knowledge without per-tool custom integrations.

External MCP coordination

Register third-party MCP servers and manage how tools enter passthrough chat from one control layer.

Why it matters: Tool wiring becomes a namespace property, not a separate project per client.

Metadata and filtering

Scope retrieval with tags, metadata, and time-aware filters — not basic vector search only.

Why it matters: Better scoping means fewer cross-domain answers and more relevant injected context.

Retrieval tuning

Adjust top-K, score thresholds, max chunks per document, context token budgets, and related retrieval behavior.

Why it matters: Serious RAG control without editing code or redeploying client apps.

Web search blending

Optionally blend fresh web results into passthrough context when configured.

Why it matters: Grounded local knowledge plus timely external facts in one pipeline.

Server stats

Observe requests, latency, chunks returned, web searches, corpus searches, passthrough chats, MCP calls, and namespace usage.

Why it matters: Debug the local context layer with numbers — not guesswork.

One local control plane for context, RAG, passthrough, and MCP

Built for people who are past the RAG demo stage.

Editions that match increasing architectural power

Pricing reflects what you need from the local context layer — not just how many files you upload. Try free, then upgrade for multiple corpora, API passthrough, MCP, metadata, and team governance.

Coming later for teams and enterprises

These capabilities are planned for future Team and Enterprise editions. They are not part of the current launch.

Honest limitations

• Retrieval quality depends on your documents, chunking, and prompts — validation is still required
• Not every edge case in RAG is handled automatically
• Performance varies with corpus size, hardware, and chosen models
• Some advanced customization may require deeper configuration

Froggy Context Hub does not claim to eliminate testing and tuning on your data. It gives you one local context hub to manage that work instead of repeating it in every client.

Frequently asked questions