INTLLM v1.0.2 · Production Release

Build a Local AI That Evolves.

Name: INTLLM
Author: Al Shahriar Sowan

INTLLM creates a continuous learning environment around local models, keeping their knowledge fresh through live information, memory, and tools.

Download for Windows ↓ Explore INTLLM →

Terminal

PowerShell

> irm https://intllm.vercel.app/install.ps1 | iex

01 / Concept

How INTLLM transforms local models

INTLLM is a local-first AI runtime built around local models served by Ollama. It is the orchestration layer around the model — not a replacement model, and not a retraining pipeline.

01 / Base

Local Model

Inference runs locally through Ollama on your machine. Model weights stay unmodified, private, and offline.

02 / Context

Internet & Tools

Opt-in DuckDuckGo retrieval and sandboxed Playwright browser access with strict security gates.

03 / Continuity

Layered Memory

Flash Brain (L0), Hot Cache (L1), and persistent Secondary Brain (L2) in PostgreSQL with pgvector.

INTLLM RUNTIME Local FastAPI Orchestrator + OpenAI-compatible API (/v1)

A more capable local AI environment

Inference Local Model

→

Routing & Index Flash Brain (L0)

→

Vector Store Layered Memory

→

Sandbox Tool Gateway

→

Attribution Live Information

02 / Agent Mode

Agent Mode for coding workflows

Agent Mode is designed primarily for coding and development workflows, giving your local model controlled access to a dedicated project workspace.

Boundary

Workspace Selection

You choose the directory. The selected workspace acts as the Agent's exclusive filesystem boundary on disk.

Security

Workspace Sandbox

Every path is resolved against the workspace. Directory traversal attempts outside the boundary are rejected with 403 Forbidden.

Control

Permission Gateway

File creation, editing, deletion, directory changes, and terminal commands require user permission. The workspace root cannot be deleted.

File Operations Workspace-scoped

Safe file reading, creation, line editing, and optional deletion within the active workspace.

Terminal Execution Monitored

Commands execute with the workspace as the working directory, subject to timeout limits and path safety verification.

Approval Model Granular

Choose Allow Once, Allow Session (for harmless repetitive tasks), or Reject for any sensitive operation.

03 / Architecture

Information flowing through the runtime

Every request traverses a single verified pipeline. Privileged actions pass through the tool gateway; memory and web context are injected dynamically before local inference executes.

Flash Brain L0

Fast micro-cache and semantic routing layer. Evaluates query confidence and freshness before performing durable vector lookups.

Hot Cache L1

Short-lived context cache with TTL invalidation. Bypasses repetitive vector computation for active conversation sessions.

Secondary Brain L2

Durable persistent knowledge store in local PostgreSQL with pgvector. Stores curated facts, workflows, lessons, and provenance.

Tool Gateway Security

Strict execution barrier. Every tool call undergoes schema validation, policy checks, approval gates, and output sanitization.

Live Information Opt-in

DuckDuckGo and SearXNG search with SSRF protection, domain trust evaluation, and clear source citation. Never enabled silently.

Local Models Ollama

Plugs into your existing Ollama library. Supports any installed model family without retraining weights or cloud telemetry.

04 / Model Management

Hardware-based recommendations

INTLLM inspects your local hardware and categorizes models into practical tiers based on parameter size and memory requirements.

Tier 01

Potato

< 3B parameters

Ultra-fast inference on integrated GPUs and low-power systems. Suitable for simple queries and constrained context windows.

Tier 02

Medium

3B–< 8B parameters

Balanced performance for typical developer workflows, code completion, and context-aware chat on consumer GPUs/CPUs.

Tier 03

High

>= 8B parameters

High-capability models designed for complex reasoning, multi-file code editing, and intensive Agent tasks with dedicated VRAM.

05 / API & CLI

Local API & CLI control

INTLLM serves an OpenAI-compatible /v1 endpoint for local integrations and ships with a unified terminal command.

Local API (/v1)

OpenAI-Compatible

GET  /v1/models
GET  /v1/models/{model}
POST /v1/chat/completions

Authenticated via INTLLM-managed scrypt-hashed API keys. Supports streaming SSE and single-request memory/web toggles.

Terminal CLI

intllm Command

intllm             # Start & launch UI
intllm --version   # Print version (1.0.2)
intllm --help      # Show command reference
intllm doctor      # Verify local dependencies

Direct CLI integration registered in your PATH upon installation across Windows, macOS, and Linux.

Get started with INTLLM v1.0.2

Production Release for Windows 10/11 x64, macOS, and Linux. PolyForm Noncommercial License 1.0.0.

Download for Windows ↓ All Downloads →

Source code is publicly available under the PolyForm Noncommercial License 1.0.0 github.com/Alshahriar-07/INTLLM →

Built by Al Shahriar Sowan Independent local-first developer infrastructure