|
*if llama.cpp supports it |
openOrchestrate is a complete Local-First MoE AI Front-End built with phpDesktop-Chrome and llama.cpp.
Not just a chat UI. An orchestration layer that:
- Routes requests intelligently
- Manages multiple GGUF models
- Preserves long-term context
- Degrades gracefully on constrained hardware
Built for people who want local AI that respects limited VRAM, limited context, and reality itself.
ββββββββββββββββββββββββββββββββββββββββββ
β Query β Router β Right Model β
β Code β CodeLlama β
β Medical β Meditron β
β General β Llama β
ββββββββββββββββββββββββββββββββββββββββββ
βββββββββββββββββββββββββββββββββββββββ
β Context full β ARCHIVE & INDEX β
β β Recall when needed β
β β Keep continuity β
βββββββββββββββββββββββββββββββββββββββ
Context limits managed deliberately, not silently dropped.
ββββββββββββββββββββββββββββββββββββββββ
β GPU: [ββββββββββ] Managed β
β CPU: [ββββββββββ] Auxiliary β
β VRAM: Predictable β β
ββββββββββββββββββββββββββββββββββββββββ
ββββββββββββββββββββββββββββββββββββ
β β No API calls β
β β No telemetry β
β β No cloud β
β β Stays on YOUR machine β
ββββββββββββββββββββββββββββββββββββ
Attach text files for analysis.
Fuel the development β’ Support more features β’ Keep it 100% local & independent
USER INTERFACE
(HTML/CSS/JS)
β
βΌ
PIPELINE ENGINE
β
ββββββββββΌβββββββββ
βΌ βΌ βΌ
LLAMA VELOCITY CONTEXT
GOVERNOR INDEX PRUNING
β
βΌ
llama.cpp
| Component | Tech |
|---|---|
| Frontend | HTML/CSS/JS (~66k chars) |
| Backend | PHP (~40k chars) |
| Runtime | phpDesktop-Chrome |
| Inference | llama.cpp |
βββββββββββββββββββββββββββββββββββββββββββββββββββββ
β IF llama.cpp CAN RUN IT, WE CAN ORCHESTRATE IT β
βββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
π£οΈ GENERAL LLMs
|
π» CODE MODELS
|
|
π₯ MEDICAL/RESEARCH
|
π― ANY GGUF
|
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β β‘ Constraints are real β
β β‘ Regression is failure β
β β‘ Working paths are sacred β
β β‘ Graceful degradation > Silent failure β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
Focus: Stability > Features | Approach: Conservative releases
| πΎ Limited VRAM |
π Limited Context |
π€ User Trust |
π Reality |
Local AI deserves tooling that respects constraints.
βββββββββββββββββββββββββββββββββββββββββββββββββββ
β β
β "Finally, a local AI tool that doesn't β
β treat me like I have a datacenter" β
β β Hopefully Youβ
β β
βββββββββββββββββββββββββββββββββββββββββββββββββββ
Issues are welcome. PRs are reviewed. Respect is expected.

