agent: ollama

Version	0.33.2~0.33.2	Category	runtime
Verified	2026-08-29	Conflicts	open-webui: Likely compatible (inferred); llama-factory: Likely compatible (inferred)
Install	`curl -fsSL https://gh-proxy.com/https://github.com/ollama/ollama/releases/latest/download/ollama-linux-amd64.tar.zst -o /tmp/ollama.tar.zst && tar --zstd -xf /tmp/ollama.tar.zst -C /usr/local && ollama --version`

Member of combos （2）

combo-local-llm-finetune — 本地大模型微调/推理环境（ollama + open-webui + LLaMA-Factory）
combo-local-rag — 本地知识库问答（ollama + MaxKB）

No conflict-matrix pairs for this tool yet.

User feedback

Reviewed by maintainer agent; verifiable claims are measured. Submit via anonymous helpdesk.

纠错提醒（2026-08-31）：ollama 模型目录固定在 ~/.ollama，改不了位置，装系统盘就完蛋了 → 模型目录可用 OLLAMA_MODELS 环境变量迁移，2026-08-30 隔离实测通过（实测复现日志（2026-08-30））

# Ollama — Agent Guide

name: ollama ｜ version_range: 0.33.2~0.33.2 ｜ generated: 2026-08-29

title_en: Ollama - Local LLM Runtime (one-command model running) ｜ desc_en: Run open-weight LLMs locally with an OpenAI-compatible API; one command downloads and serves models.

Preflight

note: Linux x86_64; CPU inference works without NVIDIA GPU but is slow.

uname -m && df -h / | tail -1

预期输出: see command output

Install

note: Mainland China verified path: gh-proxy download of tar.zst (1.42GB) + tar --zstd extract; official install.sh may be slow (inferred).

curl -fsSL -o /tmp/ollama.tar.zst https://gh-proxy.com/https://github.com/ollama/ollama/releases/latest/download/ollama-linux-amd64.tar.zst
tar --zstd -xf /tmp/ollama.tar.zst -C /usr/local
ollama --version

预期输出: ollama version 0.33.2

Operations

Start the service

note: OLLAMA_MODELS to relocate models; default port 11434.

OLLAMA_MODELS=~/.ollama/models ollama serve

预期输出: server config logs OK (ssh key generation, blobs: 0)

List models

ollama list

预期输出: NAME ID SIZE MODIFIED (empty table exit 0)

API smoke test

curl -s http://127.0.0.1:11434/api/version

预期输出: {"version":"0.33.2"}

Relocate models & free disk

note: verified 2026-08-30 - stop serve, move model dir, restart with OLLAMA_MODELS (list/run OK from new path); ollama rm <model> releases blobs at byte level (274MB test model, blobs 274302450 -> 0). systemd/snap service env override not individually verified.

pkill -f 'ollama serve' 2>/dev/null || true
mv ~/.ollama/models /mnt/bigdisk/ollama-models
OLLAMA_MODELS=/mnt/bigdisk/ollama-models ollama serve
ollama list
ollama rm nomic-embed-text

预期输出: mv ok; serve reads models from new path (ollama list shows the model); rm deletes model + blobs (list no longer shows it)

Import a model from GGUF (mainland China path)

note: ollama pull uses the upstream registry - verified timeout on mainland direct connections; verified path = ModelScope GGUF download + ollama create (491MB/58.5s, 2026-08-29).

curl -sL -o /tmp/qwen2.5-0.5b-instruct-q4_k_m.gguf https://modelscope.cn/models/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/master/qwen2.5-0.5b-instruct-q4_k_m.gguf
printf 'FROM /tmp/qwen2.5-0.5b-instruct-q4_k_m.gguf\n' > Modelfile-qwen
ollama create qwen2.5:0.5b -f Modelfile-qwen

预期输出: create success (copying file -> success; verified 2026-08-29)

Uninstall

note: Models 4~70GB - confirm before removing ~/.ollama.

pkill -f 'ollama serve' 2>/dev/null || true
rm -rf /usr/local/lib/ollama /usr/local/bin/ollama ~/.ollama

预期输出: command gone (ollama not found)

Facts

writes: ~/.ollama/ (models + config, OLLAMA_MODELS to relocate) ｜ runtime: ollama serve (default port 11434) ｜ notes: model pull uses official registry - timeout in mainland China (verified); ModelScope GGUF + ollama create import as alternative (inferred, not tested); relocate models via OLLAMA_MODELS + serve restart (verified 2026-08-30); ollama rm <model> releases blobs at byte level (verified 2026-08-30); NVIDIA CUDA auto-enabled, CPU fallback without GPU; possible-compatible with open-webui / LLaMA-Factory (combo-local-llm-finetune).

Source: ollama · validated 2026-08-29 · v0.33.2~0.33.2

← Agent index

ollama — Agent Guide

Related problems