Skip to content

[Windows] Offline installation fails: platform-specific dependencies (pyreadline3, poetry-core) not resolved #1648

@3x3mugo

Description

@3x3mugo

Describe the bug

Open Interpreter cannot be installed offline on Windows due to missing platform-specific dependencies (notably pyreadline3) and build system requirements (poetry-core).

Investigation context:

  • Time invested: 80+ hours across multiple installation methods
  • Success rate: 0% (complete failure on air-gapped system)
  • Hardware: High-end system successfully running Ollama with 70B models
  • Network status: Intentionally offline/air-gapped by design

Primary blockers:

  1. pip download --platform win_amd64 does NOT fetch Windows-only dependencies like pyreadline3 when run on non-Windows machines
  2. poetry-core not listed as a direct requirement, but required for source installs
  3. Dependency graph is complex (50+ packages, many with platform-specific binaries)

This makes Open Interpreter completely unusable for air-gapped/offline Windows environments (enterprise, government, research labs, etc.)

Reproduce

On macOS/Linux machine WITH internet:

  1. mkdir open-interpreter-offline && cd open-interpreter-offline
  2. pip download --platform win_amd64 --python-version 311 --only-binary=:all: open-interpreter -d windows_deps
  3. Result: Downloads ~30-40 packages but MISSES Windows-specific ones

Transfer to Windows machine WITHOUT internet:

  1. Copy windows_deps folder via USB
  2. cd windows_deps
  3. pip install --no-index --find-links . open-interpreter
  4. Error: "No matching distribution found for pyreadline3>=3.4.1; sys_platform == 'win32'"

Also tried - Source installation:

  1. git clone https://github.com/OpenInterpreter/open-interpreter.git
  2. cd open-interpreter
  3. pip install -e .
  4. Error: "No matching distribution found for poetry-core>=1.0.0"

All 8 attempted methods failed. See attached report for complete details.

Expected behavior

Offline installation should succeed using pre-downloaded packages if dependency resolution works correctly for the target platform.

Expected: pip install --no-index --find-links . open-interpreter should install successfully with all Windows-specific dependencies included in the download.

Actual: Installation fails because platform-specific dependencies (marked with ;sys_platform == "win32") are evaluated on the DOWNLOAD machine, not the TARGET machine. When downloading on macOS/Linux, pip ignores Windows-only requirements entirely.

Screenshots

No response

Open Interpreter version

0.0.296 (attempted installation, never successfully installed)

Python version

3.11.9

Operating System name and version

Windows 10/11 (64-bit) - Air-gapped system Successfully running: Ollama 0.12.3, DeepSeek R1:70b, other AI tools (Jan, LM Studio)

Additional context

Complete 80+ hour investigation documented including:

  • All 8 attempted installation methods with detailed error logs
  • Complete dependency chain analysis (50+ packages)
  • Time investment breakdown
  • Root cause analysis (cross-platform dependency resolution failure)
  • Comparison with alternatives (PyGPT, Agent Zero, Aider)
  • Suggested fixes prioritized (HIGH/MEDIUM/LOW)

Missing Windows-specific dependencies:

  • pyreadline3>=3.4.1 (sys_platform == "win32")
  • poetry-core>=1.0.0 (build system)
  • colorama, appdirs, and ~40+ more packages

Why cross-platform download fails:
Platform-specific conditionals (;sys_platform == "win32") are evaluated on the DOWNLOAD machine (macOS/Linux), not the TARGET machine (Windows). This causes pip to skip Windows-only packages entirely.

Suggested HIGH PRIORITY fixes:

  1. Provide platform-specific offline bundles (.zip files for Windows/Linux/macOS)
  2. Create script to properly resolve platform-specific dependencies
  3. Document Windows-specific dependencies explicitly

Full technical report: windows-offline-failure-analysis.md (attached)

Alternatives that work offline:

  • PyGPT (desktop app, 20min install, includes voice)
  • Agent Zero (Docker-based, full autonomy)
  • Aider (lighter dependencies, code editing)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions