LLM Server — On-Device AI API

magnus2026

Mayelana nalolu hlelo lokusebenza

★ Guqula ifoni yakho ibe yiseva ye-AI ★

Iseva ye-LLM ikuvumela ukuthi usebenzise amamodeli ezilimi ezinkulu ngqo kudivayisi yakho ye-Android — futhi uwaveze njengendawo yokugcina ye-API yendawo lapho noma yiluphi uhlelo lokusebenza, iskripthi, noma i-ejenti ye-AI ingaxhuma khona.

Akukho cloud. Akukho okubhaliselwe. Ifoni yakho iba yi-backend ye-AI.

━━━━━━━━━━━━━━━━━━━━━━━━━━━

🧠 ISITHELEKISO SEDIVAYISI
• Sebenzisa ama-LLM avulekile (i-Llama, i-Gemma, i-Phi, i-Qwen, i-Mistral nokuningi)
• Kwenzelwe ihadiwe yeselula nge-quantization ye-GGUF
• Ubumfihlo obugcwele — idatha yakho ayiphumi kudivayisi yakho
• Isebenza ngokuphelele ungaxhunyiwe ku-inthanethi uma amamodeli elandiwe

🔌 I-API SERVER EYAKHIWE NGOKUBHALA
• I-REST API ehambisana ne-OpenAI ekhonzwa kusuka kufoni yakho
• Xhuma noma yiliphi ithuluzi elikhuluma ifomethi ye-OpenAI API
• Sakaza izimpendulo ngesikhathi sangempela ngemicimbi ethunyelwe yi-Server
• Khonza amaklayenti amaningi kunethiwekhi yakho yendawo

🤖 KUPHELELE KUMA-AGENTI E-AI
• Vumela i-AI yakho abasizi bashayela ifoni yakho njenge-backend ye-LLM

• I-Hermes Agent, i-LangChain, i-AutoGPT — noma yini ene-HTTP
• Usekelo lwe-Tailscale/ZeroTier lokufinyelela okukude okuphephile
• Azikho izinkinobho ze-API ezidingekayo — ungumnikazi wemodeli

📱 INDLELA ESEBENZA NGAYO
1. Landa imodeli kusuka kusiphequluli semodeli esakhelwe ngaphakathi
2. Thepha okuthi "Qala Iseva" ukuze uqalise i-API endpoint
3. Khomba noma yiliphi iklayenti ku-http://your-phone-ip:8080/v1/chat/completions
4. Yilokho kuphela — ifoni yakho manje isiyiseva ye-AI

🔧 IMINININGWANE YOBUCHWEPHESHE
• i-llama.cpp backend ukuze uthole i-CPU/GPU esheshayo
• Isekela amamodeli e-GGUF (Q4_K_M, Q5_K_M, Q8_0 nokuningi)
• I-OpenAI-compatible /v1/chat/completions endpoint
• Ubude bomongo obulungisekayo, izinga lokushisa, kanye nokusampula
• Isevisi yangemuva iyaqhubeka isebenza ngisho nalapho uhlelo lokusebenza luncishisiwe
• I-inference elungiselelwe ibhethri ukuhlela

━━━━━━━━━━━━━━━━━━━━━━━━━━━━

💡 SEBENZISA AMAKESI
• Umsizi we-AI yomuntu siqu osebenza ngo-100% kudivayisi yakho
• Ukuthuthukiswa nokuhlolwa — izici ze-AI zesibonelo ngaphandle kwezindleko zamafu
• I-backend ye-agent ye-AI — vumela ama-agent azimele asebenzise ifoni yakho njengobuchopho bawo
• I-AI yokuqala yobumfihlo — idatha ebucayi ayilokothi ithinte ifu
• I-AI engaxhunyiwe ku-inthanethi — isebenza ezindizeni, emihubheni, noma kuphi

🔒 UBUMFIHLO NOKUPHEPHA
• Konke ukuqagela kwenzeka kudivayisi
• Akukho ukuqoqwa kwedatha, akukho telemetry, akukho ukuxhomekeka kwamafu
• Iseva ye-API itholakala kuphela kunethiwekhi yakho yendawo ngokuzenzakalelayo
• I-ecosystem yemodeli yomthombo ovulekile — akukho mthengisi khiya

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Iseva ye-LLM ihlanganisa igebe phakathi kwamamodeli e-AI anamandla omthombo ovulekile namadivayisi owaphatha nsuku zonke. Yeka ukukhokhela izingcingo ze-API yamafu — sebenzisa i-AI yakho, ngemibandela yakho.

Landa manje bese uguqula ifoni yakho ibe yindlu enamandla ye-AI.

Yini entsha

v1.7.0 — Improved first-launch experience with automatic starter model setup. New users get a working AI model within a minute of opening the app.