Status and roadmap
Early alpha
Everything listed here is a work in progress. "Shipped" means the code exists on PyPI and you can install it, not that it's stable or production-ready. APIs will change. This page will be updated as things solidify.
What's shipped
Core libraries (Phase 1)
| Package | Version | PyPI | Status |
|---|---|---|---|
| qwen-think | v0.1.2 | pip install qwen-think |
Shipped |
| qwen3.6-mtp | v0.1.1 | pip install qwen3.6-mtp |
Shipped |
| qwen3-repo | v0.1.0 | pip install qwen3-repo |
Shipped |
| forge-infer | v0.3.0 | pip install forge-infer |
Shipped (includes ForgeEngine) |
| qwen-compat | -- | GitHub | Test matrix + upstream PRs |
Product layers (Phase 2)
| Package | Version | PyPI | Status |
|---|---|---|---|
| forge-observe | v0.1.1 | pip install forge-observe |
Shipped |
| forge-infer-cloud | v0.1.2 | pip install forge-infer-cloud |
Shipped |
| forge-dashboard | v0.1.1 | pip install forge-dashboard |
Shipped |
| forge-studio | v0.1.0 | GitHub | Shipped (FastAPI + React) |
| Open WebUI plugins | -- | GitHub | Shipped (4 plugins) |
Upstream contributions
| PR | Repo | Status |
|---|---|---|
| #15899 | ollama/ollama | Open -- qwen35moe architecture support |
| #15901 | ollama/ollama | Open -- format constraint for all thinking parsers |
| #15902 | ollama/ollama | Open -- NVFP4 BF16 exemption for linear_attn |
What's planned
Multi-model support. The session and routing layers are designed around Qwen3.6 today. Mistral Small 4 (configurable reasoning effort), GLM-5.1, and DeepSeek V4 (dual thinking/non-thinking mode) have the same patterns. Adding support means writing backend normalizers for each model's flag format -- the router and budget manager are already model-agnostic in structure.
Domain-specific agents (Phase 3). Medical coding, legal research, deep research -- same Forge infra stack with domain-tuned routing policies. Not fine-tuning models; tuning the control plane. These only make sense to build once the core tools have real users asking for them.
Hosted deployment. forge-cloud and forge-dashboard are both designed for self-hosting today. Managed hosting is planned if there's demand.
How development is prioritized
New features and modules ship when there's demonstrated demand -- GitHub issues, community requests, or usage patterns that justify the work. If you're using any of these packages and have a use case that's not covered, open an issue on the relevant repo.