# Ollama Termux > Built from upstream [Ollama](https://github.com/ollama/ollama), adapted as a > Termux-first fork for Android ARM64 devices. [![npm](https://img.shields.io/npm/v/@mmmbuto/ollama-termux?style=flat-square&logo=npm)](https://www.npmjs.com/package/@mmmbuto/ollama-termux) [![platform](https://img.shields.io/badge/platform-Android%20ARM64-3DDC84?style=flat-square&logo=android&logoColor=white)](https://termux.dev) [![license](https://img.shields.io/badge/license-MIT-blue?style=flat-square)](./LICENSE) --- ## What This Is `ollama-termux` is an explicit fork of upstream Ollama for Termux on modern Android ARM64 phones. It keeps the upstream Ollama codebase and release naming scheme, but adds a Termux-specific distribution flow and mobile-oriented runtime behavior. ### What We Keep From Upstream - Upstream Ollama source tree and MIT license - Upstream version lineage, published as `v-termux` (or, for patch releases, `v-termux.N`) - Standard `ollama` CLI and server behavior ### What This Fork Changes - Exposes only the launcher integrations verified on Termux: **Codex**, **Codex VL**, **Qwen Code**, **Pi** - Auto-installs missing integrations from npm straight from the launcher menu - Uses `termux-open-url` for browser/OAuth flows - Tunes CPU thread selection, memory heuristics, and context limits for modern phones - Resolves Vulkan through Android's system loader before Termux Mesa, avoiding silent llvmpipe/CPU fallback - Ships prebuilt Android ARM64 release assets through GitHub Releases and installs them through the npm package ### Termux-Specific Runtime - Inference runs through the upstream `llama-server` subprocess (Ollama 0.32.x architecture), cross-built for Android ARM64 - RAM budget: never exceeds Linux `MemAvailable`; reserves Android headroom and backs off further when zram/swap is nearly exhausted - Thread limit: big cores only (cpufreq-based detection) - Context window: auto-limited based on available RAM tiers - Library paths: `/system/lib64` before `$PREFIX/lib`, wired into the `llama-server` subprocess so the vendor driver wins over Termux Mesa/llvmpipe --- ## Installation ```bash pkg update && pkg upgrade -y pkg install nodejs-lts -y npm install -g @mmmbuto/ollama-termux@latest ollama-termux # run the installer once ``` The npm package is an installer wrapper: `ollama-termux` downloads the matching GitHub Release asset, verifies SHA256, and installs `bin/ollama` + the `lib/ollama` runtime (llama-server + backend libraries) under the Termux prefix. Recent npm versions block `postinstall` scripts by default (allow-scripts), so running `ollama-termux` after the install is the reliable path; on older npm the postinstall hook does the same thing automatically. --- ## Supported Integrations | Order | CLI | Package | Status | |-------|-----|---------|--------| | 1 | **Codex VL** | `@mmmbuto/codex-vl` | Vivling-enhanced fork — **primary on Termux** | | 2 | **Codex** | `@mmmbuto/codex-cli-termux` | Termux fork | | 3 | **Qwen Code** | `@mmmbuto/qwen-code-termux` | Termux fork | | 4 | **Pi** | `@earendil-works/pi-coding-agent` | Upstream npm, Termux-compatible | On Termux, Codex VL is the primary integration: it is listed first in the menu, and a bare `ollama` / `ollama launch` on a fresh install (no prior selection) drops straight into Codex VL. If Codex VL is not installed, the menu is shown instead. The launcher offers to install a missing integration when you select it (npm-based, with confirmation). Manual install also works: ```bash # Codex VL — our Vivling fork (primary on Termux) npm install -g @mmmbuto/codex-vl # Codex — our Termux fork npm install -g @mmmbuto/codex-cli-termux # Qwen Code — our Termux fork npm install -g @mmmbuto/qwen-code-termux # Pi npm install -g @earendil-works/pi-coding-agent ``` --- ## Quickstart ```bash # Start Ollama ollama serve & # Pull recommended local models ollama pull qwen3.5:4b ollama pull gemma4:e4b # On Termux, first run drops straight into Codex VL (primary); # otherwise pick chat or a CLI from the menu ollama # Or launch an integration directly ollama launch codex-vl --model gemma4:e4b ollama launch codex --model qwen3.5:4b ollama launch qwen --model qwen3.5:4b ollama launch pi ``` ### Larger phone models `ornith:9b` can be pulled by this Ollama base and is the largest Ornith variant worth testing on a 16 GB phone. Its advertised 256K context is not a mobile target: the fork clamps context from the conservative live-memory budget. Start with the default/4K context and `OLLAMA_VULKAN=1`; stop if Android begins heavy zram swapping. `ornith:35b` is not a supported phone target. --- ## Build ```bash export NDK_ROOT=~/android-ndk/android-ndk-r27c ./scripts/build_termux.sh ``` Output: `dist/ollama-termux--android-arm64.tar.gz` The upstream project lists integrations such as [Claude Code](https://docs.ollama.com/integrations/claude-code), [Codex](https://docs.ollama.com/integrations/codex), [Copilot CLI](https://docs.ollama.com/integrations/copilot-cli), [DeepSeek Harness](https://docs.ollama.com/integrations/deepseek-harness), [Droid](https://docs.ollama.com/integrations/droid), and [OpenCode](https://docs.ollama.com/integrations/opencode); on Termux this build supports only the launcher-verified set above: **Codex**, **Codex VL**, **Qwen Code**, **Pi** — the other integrations are refused by the launcher and are not shipped in the Termux package. See [docs/BUILDING.md](./docs/BUILDING.md) for the full cross-build flow. --- ## Devices Runtime-validated device: - Pixel 9 Pro / Tensor G4 / Mali-G715 (CPU and Android-system-loader Vulkan) Additional ARM64 targets covered by runtime-dispatched CPU variants and the Android Vulkan path, but still requiring release-candidate device validation: - ASUS ROG Phone 3 / Snapdragon 865+ / Adreno 650 - Galaxy S24+ / Snapdragon 8 Gen 3 / Adreno - Galaxy S25 Ultra / Snapdragon 8 Elite / Adreno CPU backends: runtime-dispatched llama.cpp variant libraries (`GGML_CPU_ALL_VARIANTS`), selected per device at startup. Optional Vulkan GPU backend (`BUILD_VULKAN=1`). --- ## Links - npm: https://www.npmjs.com/package/@mmmbuto/ollama-termux - Releases: https://github.com/DioNanos/ollama-termux/releases - Upstream: https://github.com/ollama/ollama --- ## License MIT — original upstream [ollama/ollama](https://github.com/ollama/ollama). Termux fork work: DioNanos. --- ## Contact Maintained by [DioNanos](https://github.com/DioNanos). - General / dev: **dev@mmmbuto.com** - Security disclosures: **security@mmmbuto.com** - Project hub: