Repository navigation

Release list

IntLLM v1.1.0 Latest

Latest

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):

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

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

IntLLM v1.0.2

Alshahriar-07 released this 02 Oct 11:31

v1.0.2

0a7823d

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.

Version: 1.0.2
Release status: Production release (Windows x64, Python package, source distribution)
Release date: 2026-10-02
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)	Standalone executable + installer
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.exe`	Windows x64	Standalone application (PyInstaller single-file)
`INTLLM-Setup.exe`	Windows x64	Windows installer (Inno Setup)
`intllm-1.0.2-py3-none-any.whl`	Python	PyPI / package installation
`intllm-1.0.2.tar.gz`	Python	Source distribution
`SHA256.txt`	All	Artifact verification
`install.ps1`	Windows	Automated installation
`install.sh`	Unix-like	Automated installation

Installation

Windows (PowerShell)

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

Or run INTLLM-Setup.exe directly.

Linux / macOS

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

Python package

python -m pip install intllm

CLI commands

intllm                 start the local runtime and open the web UI
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: 1.0.2
Publishing: PyPI Trusted Publishing (OIDC) via .github/workflows/pypi.yml
Artifacts: intllm-1.0.2-py3-none-any.whl, intllm-1.0.2.tar.gz

Requirements

Ollama installed and running with at least one model pulled
(https://ollama.com). Not bundled.
PostgreSQL 14+ with the pgvector extension. Not bundled.
Default URL: postgresql+asyncpg://intllm:intllm@127.0.0.1:5432/intllm
(INTLLM_DATABASE_URL).
Python 3.11+ for the wheel/CLI path.

INTLLM detects these dependencies and reports their state honestly
(connected / degraded / unavailable / offline).

API

Local OpenAI-compatible API:

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

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.

Agent Mode

Agent Mode adds workspace-scoped coding capabilities to Chat:

Workspace — the selected folder is the Agent's filesystem boundary.
Sandbox — every path is resolved against the workspace and escapes are
rejected (403). The workspace root cannot be deleted.
Permissions — overwrite, delete, move and terminal execution require an
explicit approval decision; allow_session grants suppress repeat prompts for
harmless repeated work.
Terminal — commands run through the backend with the workspace as the
working directory and a configurable timeout; commands referencing paths
outside the workspace are flagged.

Known limitations

Ollama and PostgreSQL are external and must be installed by the user; INTLLM
does not bundle or silently install them.
/v1/embeddings, /v1/responses, audio, images, files, batches and
fine-tuning are not implemented.
Model-driven autonomous tool-calling in Agent Mode is not yet wired; the
Agent runtime is driven through the Agent API / UI.
A browsable public model catalog is not provided; the model list is the real
Ollama inventory.
No third-party CLI compatibility is claimed unless verified against a build.
The Windows installer was compiled and statically validated (PE header, version
metadata, payload and license page), but a live install/uninstall was not
executed in this environment because it would modify the current user's PATH
and per-user locations. Run INTLLM-Setup.exe interactively, or
INTLLM-Setup.exe /VERYSILENT /SUPPRESSMSGBOXES /NORESTART, to install.

Testing status

Check	Result
Backend unit tests (`pytest`)	80 passed, 6 skipped (DB integration tests skip without `INTLLM_TEST_DATABASE_URL`)
Frontend typecheck (`tsc --noEmit`)	passed
Frontend production build (`vite build`)	passed
Windows executable build (`build_windows.py`)	built — 34.0 MB, x64 (AMD64), icon + version metadata
Executable smoke test (`/api/health`, SPA fallback)	passed
Executable runtime checks (`/`, `/api/health`, `/api/models`, `/api/agent/status`, `/chat`, `/v1/models`)	passed (real Ollama inventory; honest `503 degraded` with PostgreSQL down)
Executable CLI (`INTLLM.exe --version` / `--help`)	passed (`INTLLM 1.0.2`)
Wheel + sdist build	built; `twine check` PASSED
Wheel install + console script (`intllm --version` / `--help`)	passed
sdist independent rebuild	passed (builds both wheel and sdist)
Windows installer (`INTLLM-Setup.exe`)	built with Inno Setup 6, version metadata + payload verified; install/uninstall not executed at runtime (would modify the user's PATH and profile)
SHA256 (`SHA256.txt`)	generated from the real artifacts and re-verified independently
PostgreSQL / pgvector live integration	not available in this environment (reported honestly at runtime)
Ollama live integration	detected and reported correctly (models listed)

See the Stage 2 release report for exact commands and outputs.

License

PolyForm Noncommercial License 1.0.0

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

IntLLM v1.0.1

Alshahriar-07 released this 01 Oct 16:44

v1.0.1

91392e5

INTLLM Release Notes

Version 0.4.2

Version: 0.4.2 (single source of truth: backend/app/__init__.py)
Release date: 2026-10-01
Commit range: production backend, local API, database, documentation,
installers and release engineering.

Highlights

A production-ready, local-first runtime with an OpenAI-compatible API at
/v1, an INTLLM-owned API key system, and an in-app Docs page.
One-command builds that produce a Python wheel, source archive, a Windows x64
executable and a SHA256 manifest, driven by tested CI/CD workflows.
Installers for Windows (install.ps1) and Linux/macOS (install.sh) that
download, verify and install the correct release artifact.

Backend changes

CLI added (app.cli) with --version, --help, start, serve, doctor.
Windows launcher gained --version / --help and a raised health timeout
(10 s) so the web UI opens even when Ollama is unreachable and /api/health
is slow.
Repaired a corrupted SSE generator in app/api/routes/openai.py.
Removed dead code in the Ollama status service.

API changes

Implemented endpoints (documented and test-covered):
- GET /v1/models
- GET /v1/models/{model}
- POST /v1/chat/completions — stream: false and stream: true (SSE)
Authentication via Authorization: Bearer <key> or x-api-key.
Extra request fields intllm_use_brain / intllm_use_web control memory and
live retrieval for a single request.
Not implemented and deliberately undocumented: /v1/embeddings,
/v1/responses, audio, images, files, batches, fine-tuning.

Database changes

Schema managed by Alembic (backend/migrations, current revision 0001) with
a SQLAlchemy metadata fallback for the frozen executable.
Startup verifies PostgreSQL reachability, the pgvector extension and schema
state; failures are reported as unavailable / misconfigured with an
actionable step.
New opt-in integration tests exercise migrations, persistence and restart
recovery against a real database.

Frontend compatibility

New Docs page and a consolidated API settings section.
Existing services, tabs and design tokens are unchanged; NavigationTab
gained a docs member.
The frontend never receives or renders stored secrets (only a masked
fingerprint after creation).

Installer changes

IRM_INSTALL/install.ps1: architecture check (x64), latest/pinned version
resolution, SHA256 verification, per-user install, PATH update, Start Menu
shortcut, --version verification.
IRM_INSTALL/install.sh: POSIX shell, Python 3.11+ check, SHA256
verification, isolated venv install, intllm wrapper, verification.
Both are copied verbatim to public/ (and served as /install.ps1,
/install.sh); IRM_INSTALL/ remains the single source.
IRM_INSTALL/INTLLM-Setup.iss defines the per-user Inno Setup installer.

Packaging

Artifacts are staged in build/release/:

build/release/
├── INTLLM-windows-x64.exe
├── INTLLM-Setup.exe                 # when Inno Setup is available
├── intllm-0.4.2-py3-none-any.whl
├── intllm-0.4.2.tar.gz
└── SHA256.txt

dist/ is reserved for the Vite frontend build (which also carries the
served installer scripts), so it is not the release staging directory.

The Windows executable is a PyInstaller onefile build bundling the backend and
the built frontend. Compute the wheel/sdist checksums with
python scripts/make_sha256.py build/release.

Security changes

API keys: salted scrypt hash + short fingerprint only; full key shown once.
Loopback bootstrap access ends as soon as a key exists.
Secret redaction in logs; no secrets in bundles, Git or artifacts.
Localhost binding by default; restricted CORS; request-size limits.
Tool execution always passes through the permission gateway.

Breaking changes

The Python distribution is renamed from intllm-backend to intllm, and
the CLI entry point is now intllm (previously intllm pointed directly at
the launcher). Console script intllm-api is unchanged.
The Windows executable is named INTLLM-windows-x64.exe (previously
INTLLM.exe).
INTLLM_PORT must be a valid TCP port (0–65535).

Migration requirements

No manual database steps: run INTLLM and the schema is applied automatically
(Alembic where available, metadata DDL otherwise).
Existing deployments using a intllm-backend install should install the
intllm package instead; user data under the data directory is untouched.

Known limitations

PostgreSQL and Ollama are not bundled or auto-installed. They are external
local dependencies that INTLLM detects and reports.
Embeddings / responses / audio / images / files are not implemented.
No third-party CLI compatibility is claimed unless verified against a
build.
Background learning performs low-priority memory/index maintenance; it does
not continuously retrain Ollama model weights.
INTLLM-Setup.exe is produced only where Inno Setup's compiler (iscc) is
available; the release workflow installs it on the Windows runner. The
Linux/macOS path installs the universal wheel.
The generated INTLLM-Setup.exe was compiled and statically validated (PE
header, Inno Setup payload, SHA256) but not executed in this environment,
because a silent test install would modify the user PATH and per-user
locations. Run it interactively, or INTLLM-Setup.exe /VERYSILENT /SUPPRESSMSGBOXES /NORESTART, to install.

Verification results (this build)

Check	Result
Backend unit tests (`pytest`)	53 passed, 6 skipped
Database integration tests	skipped locally (no `INTLLM_TEST_DATABASE_URL`); run in CI with a pgvector PostgreSQL service
Backend lint (`ruff`, E/F/I)	passed
Frontend typecheck (`tsc --noEmit`)	passed
Frontend production build (`vite build`)	passed
Wheel build (`intllm-0.4.2-py3-none-any.whl`)	built
Source archive (`intllm-0.4.2.tar.gz`)	built
Wheel install + `intllm --version` / `--help`	passed (`validate_release.py --install-check`)
Windows x64 executable	built — 33.9 MB, PE machine `x64 (AMD64)`
Executable smoke test	passed — `/api/health` OK, SPA fallback OK
`INTLLM-Setup.exe`	built with Inno Setup 6 (valid PE bootstrapper, x64-gated, 35 MB); compiled and statically checked — see note below
SHA256 manifest	generated and verified for the staged artifacts
PostgreSQL / Ollama live integration	not available in this environment; reported honestly at runtime

SHA256 verification

After downloading artifacts, verify against the release SHA256.txt:

# Linux/macOS
sha256sum -c SHA256.txt

# Windows PowerShell
Get-FileHash .\INTLLM-windows-x64.exe -Algorithm SHA256

The checksums are generated by scripts/make_sha256.py and re-verified by the
release workflow (scripts/validate_release.py) before publishing. Hashes are
never typed by hand.

Artifact list

INTLLM-windows-x64.exe
INTLLM-Setup.exe
intllm-0.4.2-py3-none-any.whl
intllm-0.4.2.tar.gz
SHA256.txt

Assets 7

Releases: Alshahriar-07/INTLLM

Release list

IntLLM v1.1.0

INTLLM Release Information

Product

Supported platforms

Artifacts

Installation

Windows (PowerShell)

Linux / macOS

Python package

Desktop application behavior

CLI commands

PyPI information

Requirements

API

Agent Mode

Known limitations

Testing status (this release)

License

Uh oh!

IntLLM v1.0.2

INTLLM Release Information

Product

Supported platforms

Artifacts

Installation

Windows (PowerShell)

Linux / macOS

Python package

CLI commands

PyPI information

Requirements

API

Agent Mode

Known limitations

Testing status

License

Uh oh!

IntLLM v1.0.1

INTLLM Release Notes

Version 0.4.2

Highlights

Backend changes

API changes

Database changes

Frontend compatibility

Installer changes

Packaging

Security changes

Breaking changes

Migration requirements

Known limitations

Verification results (this build)

SHA256 verification

Artifact list

Uh oh!