Skip to content

IntLLM v1.1.0

Latest
Latest

Choose a tag to compare

View all tags
@Alshahriar-07 Alshahriar-07 released this 03 Oct 16:50
v1.1.0
74bfe95

INTLLM Release Information

Product

INTLLM — a local-first AI runtime for Ollama models with layered memory, live
retrieval, controlled tools, an Agent mode for coding work, and an
OpenAI-compatible local API, delivered as a native Windows desktop application.

  • Version: 1.1.0 (single source of truth: backend/app/__init__.py)
  • Release status: Production release — packaged application built and
    validated end-to-end against a live PostgreSQL (pgvector) and a live Ollama
  • Release date: 2026-10-03
  • Website: https://intllm.vercel.app
  • Repository: https://github.com/Alshahriar-07/INTLLM
  • License: PolyForm Noncommercial License 1.0.0
  • Copyright: Copyright © 2026 Al Shahriar Sowan

Supported platforms

Platform Support
Windows 10/11 x64 (AMD64) Portable desktop executable + installer (native window, WebView2)
Linux (x86_64 / aarch64) Python wheel (install.sh)
macOS (x86_64 / arm64) Python wheel (install.sh)

Python 3.11+ is required for the wheel/CLI path.

Artifacts

All artifacts are staged in build/release/ and published to the GitHub Release.

Artifact Platform Purpose
INTLLM-v1.1.0-win64x.exe Windows x64 Portable desktop application (PyInstaller single-file)
INTLLM-v1.1.0-Setup.exe Windows x64 Windows installer (Inno Setup 6)
intllm-1.1.0-py3-none-any.whl Python PyPI / package installation
intllm-1.1.0.tar.gz Python Source distribution
install.ps1 Windows Automated installation
install.sh Unix-like Automated installation
SHA256.txt All SHA-256 checksums of every artifact above

Installation

Windows (PowerShell)

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

Or run INTLLM-v1.1.0-Setup.exe directly, or use the portable
INTLLM-v1.1.0-win64x.exe with no installation.

Linux / macOS

curl -fsSL https://intllm.vercel.app/install.sh | bash

Python package

python -m pip install intllm

Desktop application behavior

Launching INTLLM-v1.1.0-win64x.exe (or the installed INTLLM shortcut):

  1. starts the local FastAPI backend on a real localhost TCP port
    (127.0.0.1, conflict-resolved, never a browser tab);
  2. provisions/detects local PostgreSQL — including the INTLLM-managed runtime
    (bundled binaries, loopback-only) — and applies the schema/migrations;
  3. starts the installed Ollama daemon when it is not running and detects the
    models that are actually installed;
  4. shows live readiness (PostgreSQL / Ollama / backend) with a working Retry,
    then loads the INTLLM UI into a native window; and
  5. shuts the backend and any managed PostgreSQL cluster it owns down cleanly
    when the window closes.

The root URL serves the application shell (SPA); /api/*, /v1/* and
/health//ready remain pure JSON API routes.

CLI commands

intllm                 start the local runtime and open the desktop app
intllm start           same as above
intllm serve           run the API server only
intllm doctor          check PostgreSQL, pgvector and Ollama readiness
intllm --version       print the version
intllm --help          print help

PyPI information

  • Package name: intllm
  • Version: 1.1.0
  • Publishing: PyPI Trusted Publishing (OIDC) via .github/workflows/pypi.yml
  • Artifacts: intllm-1.1.0-py3-none-any.whl, intllm-1.1.0.tar.gz

Requirements

  • Ollama installed with at least one model pulled (https://ollama.com).
    Not bundled. INTLLM starts the daemon when installed but not running.
  • PostgreSQL 14+ with the pgvector extension. When no local server is
    reachable, INTLLM uses its own managed local PostgreSQL runtime when its
    binaries are available (packaged build); otherwise it reports exactly what
    is missing. Default URL: postgresql+asyncpg://intllm:intllm@127.0.0.1:5432/intllm
    (INTLLM_DATABASE_URL).
  • Python 3.11+ for the wheel/CLI path.

API

Local OpenAI-compatible API:

GET  /v1/models
GET  /v1/models/{model}
POST /v1/chat/completions     # stream: false | true (SSE)

/v1/models returns the real Ollama inventory. /v1/chat/completions
validates the requested model, resolves it against Ollama and returns an
OpenAI-compatible response. Authenticate with Authorization: Bearer intllm_…
or x-api-key: intllm_…. API keys are created in Settings → API; only a salted
scrypt hash and a fingerprint are stored, and the full key is shown once.
Access is loopback-only by default; LAN exposure is an explicit opt-in that
still requires a key.

Agent Mode

Agent Mode adds workspace-scoped coding capabilities to Chat:

  • Workspace — the selected folder is the Agent's filesystem boundary
    (native folder picker with manual-path fallback, persisted locally).
  • Real tool loop — the local model proposes JSON tool calls
    (fs.list, fs.read, fs.search, fs.write, fs.edit, fs.mkdir,
    fs.delete, fs.move, terminal); INTLLM validates each call, applies the
    permission gate, executes it for real and feeds the result back until the
    model produces a final answer. Verified live end-to-end with
    qwen2.5-coder:7b.
  • Sandbox — every path is resolved against the workspace; traversal,
    absolute paths, drive and UNC escapes are rejected (403). The workspace
    root cannot be deleted.
  • Permissions — Allow / Ask Me. In Ask Me, workspace-changing operations
    (including terminal) pause for an explicit Allow/Deny. Reads run freely.
  • Terminal — real async subprocess execution with the workspace as cwd,
    stdout/stderr capture, exit code, duration, timeout, cancellation and
    output tailing.
  • Bounded context — file reads are size-limited (binary files return
    metadata only), search skips .git/node_modules/etc., terminal output is
    tailed, and tool transcripts are truncated before re-entering the model.

Known limitations

  • Ollama is external and must be installed by the user; INTLLM does not bundle
    or silently install it (it does start it when present).
  • /v1/embeddings, /v1/responses, audio, images, files, batches and
    fine-tuning are not implemented.
  • Agent permission-mode persistence and workspace persistence are stored in
    local PostgreSQL (app_settings) and require the database to be ready.
  • No third-party CLI compatibility is claimed unless verified against a build.

Testing status (this release)

Check Result
Backend unit tests (pytest) 131 passed, 8 skipped (DB integration tests skip without INTLLM_TEST_DATABASE_URL)
Backend lint (ruff check) 0 errors, 0 warnings (66 style findings are informational; no F/E9 class errors)
Frontend typecheck (tsc --noEmit) passed
Frontend production build (vite build) passed (code-split: app chunk ~111 kB, vendor chunks cached separately)
Windows executable build (build_windows.py) built — 50.4 MB, x64 (AMD64), icon + version metadata
Packaged EXE end-to-end health ok with PostgreSQL connected, memory connected, Ollama connected; / and /app/ serve the SPA; /api/models returned the real installed inventory (gemma3:4b, qwen2.5-coder:7b); chat streamed a real Ollama completion and persisted it to PostgreSQL
Live model registry sync passed (real Ollama metadata incl. 3.1–4.5 GB sizes, fixed via BIGINT migration)
Live chat pipeline (dev backend) passed (SSE stream → PostgreSQL history; conversations survive backend restart)
Live Agent run (real model) passed (read file → write file → final answer in 23 s; file verified on disk)
Agent terminal passed (python --version → real stdout, exit 0, 262 ms)
Workspace sandbox passed (../secret.txt and absolute C:\Users\... both rejected)
Memory store passed (create + search + stats via PostgreSQL/pgvector; survives restart)
OpenAI-compatible API passed (/v1/models real inventory; /v1/chat/completions real completion from gemma3:4b with key auth)
Wheel + sdist build built; wheel installs into a clean venv; intllm --version / --help pass
Windows installer (INTLLM-v1.1.0-Setup.exe) built with Inno Setup 6 (successful compile); interactive install/uninstall not executed in this environment because it would modify the current user's PATH and per-user locations — run it interactively or with /VERYSILENT /SUPPRESSMSGBOXES /NORESTART
SHA256 (SHA256.txt) generated from the real artifacts and re-verified independently (6/6 match)
Ollama live integration detected and exercised for real (version 0.34.2; generation verified)

License

PolyForm Noncommercial License 1.0.0

Copyright © 2026 Al Shahriar Sowan

Licensed under the PolyForm Noncommercial License 1.0.0. This is a
source-available, noncommercial license; commercial use is not permitted under
these terms. The full license text is in LICENSE.

Assets 7
Loading