Services

A single server. No subscriptions. No data leaving your machine. Eight tools for privacy-first AI.

What You Get

Service	What It Does
AI Chat Interface	A chat experience comparable to commercial tools, designed to encourage critical thinking rather than passive use
Voice Interaction	Speak to the system, hear responses. Fully local speech-to-text and text-to-speech
Research and Notes	An AI-powered environment for reading, annotating, and synthesising sources. Transforms research into podcasts
Image Generation	Generate images for presentations, projects, and creative work. Multiple model tiers from fast drafts to high quality
API Access	For developers and power users building their own tools and integrations. OpenAI-compatible endpoints
Embedded Unit Assistants	AI support embedded in platforms like Blackboard, tailored to specific document collections and knowledge bases
Rehearsal Environments	Simulated professional organisations where students practise audits, interviews, negotiations, and client interactions before the stakes are real
Coding Assistance	Connect your editor to the local LLM. VS Code with Continue, CLI tools like OpenCode, or Claude Code with local models

AI Chat

The primary interface. Users interact with local LLMs through Open WebUI — a full-featured chat interface with conversation history, multi-model switching, voice input/output, image generation, and web search. Comparable to commercial tools, running entirely on local hardware.

For research-focused work, Perplexica provides Perplexity-style answers with inline source citations and an academic search mode that prioritises scholarly sources.

Voice

Fully local speech-to-text (Whisper) and text-to-speech (Kokoro) via Speaches. Integrated natively into Open WebUI — speak, and the system listens, processes, and responds audibly. No audio data leaves the machine.

Research and Notes

Open Notebook AI is a research assistant that ingests PDFs, links, YouTube videos, and text documents. Annotate, synthesise, and generate structured summaries. The platform can transform research notes into podcasts — useful for revision and accessibility.

Image Generation

ComfyUI provides local image generation. Simple requests go through Open WebUI’s in-chat image generation (users never need to open ComfyUI directly). Users who want node-based workflow control, ControlNet, LoRAs, or advanced pipelines access ComfyUI directly in the browser.

Unit Assistants

AnythingLLM powers domain-specific RAG chatbots that can be embedded in platforms like Blackboard. Each workspace has its own document collection — a project assistant only knows that project’s content. Users ask questions about documents, procedures, and reference material and get answers grounded in the unit’s actual documents.

Rehearsal Environments

Powered by LocoEnsayo, these are AI-populated organisations that users can interrogate, interview, audit, and negotiate with. Each persona has a backstory, a role, and information constraints. The scenario is not a quiz. It is a rehearsal.

Currently deployed: CloudCore Networks (IT services firm for security audits and systems analysis) and Pinnacle Tours (hospitality and tourism).

Coding Assistance

The same Ollama instances that power the chat interface are accessible from your code editor via OpenAI-compatible API endpoints. No separate setup needed — point your tool at the local Ollama URL and start coding.

Editor integrations:

Continue — VS Code / JetBrains extension. Autocomplete, chat, and inline editing backed by local models. The most mature option.
OpenCode — Terminal-based coding assistant. Lightweight, fast, connects to any OpenAI-compatible endpoint.
Claude Code — Anthropic’s CLI tool. Can be configured to use local models via OpenAI-compatible endpoints for privacy-sensitive work.
Claude Desktop — Can connect to local Ollama via MCP (Model Context Protocol) for a desktop AI assistant backed by your local stack.

Agentic coding tools (tools that autonomously write, run, and iterate on code in loops) are technically possible with local models. We recommend caution.

Agentic drift is the compounding of errors across multi-step loops. A model that is 90% accurate per turn is only 59% accurate over five turns. Small local models amplify this — they are less reliable per step than frontier models, and the errors compound faster.

More fundamentally, agentic delegation bypasses the thinking that makes AI useful for learning. You get code in a folder; you don’t get understanding in your head. The LocoPuente stack is designed for conversation, not delegation — AI that amplifies your thinking rather than replacing it. See the Philosophy page for more on this distinction.

Use local models as a coding conversation partner: explain your approach, ask it to review your logic, brainstorm alternatives, generate test cases. That interaction makes you a better developer. Handing it autonomous control does not.

Productivity Tools

Beyond AI services, LocoPuente includes tools for everyday work:

Stirling PDF — merge, split, compress, OCR, annotate, and convert PDFs. Everyone needs PDF tools; no data leaves the machine.
Excalidraw — collaborative whiteboard and diagramming. Group projects, case study mapping, business process diagrams.
CiteSight — citation verification and writing quality checker. Pre-submission check that catches citation errors and AI-generated writing patterns before they cost marks.