LOPI Project

Local-first AI assistant
& bounded agent.

LOPI is a self-hosted AI helper that runs entirely on your own hardware — no cloud AI service, no third-party agent framework. It owns its inference runtime, its memory, and its tools, and it can carry out approved, sandboxed tasks under strict, server-enforced limits.

0
Cloud dependencies
4
Local GGUF models
200
Passing tests
12
Milestone roadmap

A capable assistant that stays in your control

LOPI blends a friendly chat experience with a carefully bounded agent layer. It can search the live web, remember what matters, read your approved documents, and act — but only within limits the server owns and you approve.

Chat

Streaming conversation on local models.

Citation-aware answers, current-date awareness, echo repair, and a one-click "Continue" for length-stopped replies — all on local inference with no cloud APIs.

Search

Grounded answers from live sources and local documents.

Controller-owned search prioritises live sources for changing questions and approved local documents for stable ones. Provider keys never touch the browser.

Memory

Local memory with review-gated document storage.

Local SQLite/FTS5 memory plus a separate document store with immutable versions, conflict handling, and rollback.

Agent

Bounded agent tasks with explicit opt-in.

One validated action at a time, a step limit, a repetition guard, and durable, replayable task events — all within server-enforced limits.

Audit

Every mutation confirmed and recorded.

Mutations pause for exact-action approval. Every task is recorded with redacted, ordered events. Restart never replays an unproven change.

One owner per responsibility

LOPI's defining trait is disciplined ownership. Each concern has exactly one authoritative component, and the browser never holds business logic, credentials, or tool execution.

YOUR MACHINE · LOOPBACK ONLY Focus UI LOPI Controller routing · policy · tools Inference Node pinned llama.cpp Control Plane tasks · confirm · audit Memory SQLite / FTS5 Knowledge approved documents

One machine. One stack. Every layer owned by LOPI — nothing leaves the device unless you allow it.

Local by default Pinned llama.cpp runtime, local models, and local databases. Works fully offline.
Code authorizes, models propose The model suggests an action; LOPI validates, permits, executes, and records it.
No cloud AI dependency No Ollama, no hosted model API, and no third-party agent framework in the runtime.
No unrestricted tools No generic shell, filesystem, or database access — only narrow, confirmed operations.

LOPI runs several local GGUF models behind node-aware aliases and independent chat, decision, and final roles. Every model is revision-pinned, license-tagged, and paired with its own instruction profile.

ModelSizeRole / useRuntimeStatus
Qwen3 1.7B (fast) 1.3 GB chat, decision, final llama.cpp b10488 Default — tuned 6/8 threads
Gemma 3 4B (balanced) 2.5 GB established CPU middle llama.cpp b10488 Own sampling profile
Qwen3 8B (quality) 5.0 GB slower, higher quality llama.cpp b10488 Temperature capped via eval
Qwen3.5 4B (qwen35) 2.7 GB experimental llama.cpp b10488 Pending fixed evaluation

Rollback runtime b10326 remains hash-locked. Sampling values are model-owned and changed only through a fixed evaluation — never copied between model families.

A careful path to autonomy

LOPI's roadmap expands capability only after the safety substrate beneath it is proven. Foundations are complete; the durable task runtime is live; sandboxed workflows come next.

Foundations Complete

Strict action protocol, bounded orchestrator, canonical tool registry, multi-model routing, agent streaming, control plane, and local knowledge store.

Durable task runtime Implemented

Owner-bound ordered task events, a detached runner with a desktop queue, replay/polling, routed cancellation, and restart-safe semantics.

Tool contracts In progress

Bounded output schemas and risk/idempotency metadata on the canonical registry — classifying existing tools before expanding the surface.

Skill Packs & HPE/LAN Planned

Administration-approved, non-executable instruction packs, then versioned least-privilege connectors to a separate HPE deployment.

Why it matters

AI assistance without giving up ownership

LOPI shows that a genuinely useful assistant — search, memory, documents, and bounded autonomous tasks — can run entirely on your own hardware, with every permission explicit and every action auditable.

Get in touch →