ollama-dl

An advanced CLI for managing Ollama models - download, inspect, verify, archive, and batch-install.

---

Build

go mod tidy
go build -o ollama-dl .

Cross-compile (no CGO needed):

GOOS=linux  GOARCH=amd64  go build -o ollama-dl-linux-amd64  .
GOOS=linux  GOARCH=arm64  go build -o ollama-dl-linux-arm64  .
GOOS=darwin GOARCH=arm64  go build -o ollama-dl-darwin-arm64 .
GOOS=windows GOARCH=amd64 go build -o ollama-dl.exe          .

---

Commands

urls — Direct download links

ollama-dl urls llama3:8b
ollama-dl urls --json deepseek-coder-v2:latest
ollama-dl urls --registry https://my-registry.local mymodel:v1

Prints HTTPS URLs for every blob, the manifest JSON, and ready-to-use wget -c commands.

---

pull — Built-in parallel downloader with progress bars

ollama-dl pull llama3:8b
ollama-dl pull -o ./downloads -j 5 deepseek-coder-v2:latest

• Resumable downloads (sends Range header, continues from partial .part files)
• Live per-blob progress bars with ETA and speed (via mpb)
• Configurable concurrency and retry count
• Saves manifest.json alongside the blobs automatically

Flags:

• -o, --out <dir> — output directory (default: <repo>-<tag>)
• -j, --concurrency <n> — parallel downloads (default: 3)
• --retries <n> — max retries per blob (default: 3)
• --registry <url> — override registry base URL

---

install — Install into local Ollama store

ollama-dl install ./llama3-8b my-llama:local
ollama-dl install ./model --symlink my-llama:local   # symlink blobs (saves disk)
ollama-dl install ./model --ollama-dir /custom/path my-llama:local

Copies (or symlinks) all blobs to ~/.ollama/models/blobs/ and registers the manifest. The model is immediately available to ollama run — no restart needed.

---

info — Model architecture & metadata

ollama-dl info llama3:8b
ollama-dl info deepseek-coder-v2:16b-lite-instruct-q4_K_M

Downloads and parses the config blob, displaying:

  Model Family          llama
  Architecture          llama
  Parameters            8B
  Format                gguf
  Quantization          Q4_0
  Context Length        8192 tokens
  Stop Tokens           <|start_header_id|> <|end_header_id|> <|eot_id|>

  Blobs:                5
  Total size:           4.7 GB

  Layers:
    model               4.7 GB
    template            1.4 KB
    params              96 B

---

diff — Compare two model versions

ollama-dl diff llama3:8b llama3:8b-instruct-q4_K_M
ollama-dl diff mistral:7b mistral:7b-instruct-v0.2

Shows exactly which blobs changed between two versions, how much data you'd need to re-download, and how much is shared (already on disk).

  + model               4.4 GB    sha256:3c37...
  - model               4.7 GB    sha256:6a0a...

  Shared blobs    : 3 (no download needed)
  Added in B      : 1  (+4.4 GB)
  Removed from A  : 1  (-4.7 GB)
  Net change: -307.2 MB

---

verify — SHA256 integrity check

ollama-dl verify                       # all installed models
ollama-dl verify llama3:8b             # single model
ollama-dl verify --fix llama3:8b       # remove corrupt blobs

Reads every installed model, SHA256-hashes each blob on disk, and compares it against the digest in the manifest. Use --fix to delete corrupt blobs so Ollama can re-pull them cleanly.

---

search — Browse the Ollama library

ollama-dl search llama
ollama-dl search "code generation"
ollama-dl search --json mistral

Queries the Ollama library API and lists matching models with their tags and pull counts. Use the result names directly with pull or urls.

---

pack — Create a portable archive

ollama-dl pack llama3:8b
ollama-dl pack llama3:8b -o llama3-8b-backup.tar.zst
ollama-dl pack llama3:8b --compression 9    # slower but smaller

Bundles the model's blobs + manifest into a single .tar.zst file. Ideal for:

• Air-gapped machine transfers (copy one file over USB/SCP)
• Long-term model archival
• Sharing specific quantizations with teammates

The archive is self-contained — no network access needed to install from it.

---

unpack — Install from an archive

ollama-dl unpack llama3-8b-20240115.tar.zst --name llama3:local
ollama-dl unpack archive.tar.zst --inspect      # list contents only

Extracts the archive and installs it into the Ollama store, exactly like install.

---

batch — Download multiple models at once

ollama-dl batch models.yaml
ollama-dl batch models.yaml --dry-run       # preview without downloading
ollama-dl batch models.yaml -j 5            # 5 parallel downloads per model

Reads a YAML (or JSON) manifest:

out_dir: ./downloaded-models

models:
  - model: llama3:8b
    auto_install: true

  - model: mistral:7b-instruct-v0.2-q5_K_M
    name: mistral-q5:local        # install under this name
    auto_install: true

  - model: deepseek-coder-v2:latest
    auto_install: false            # download only, install later

  # Private registry support
  - model: myorg/custom-model:v2
    registry: https://my-registry.example.com
    auto_install: true

Blobs shared across models are downloaded only once and symlinked for the rest (deduplication).

---

Environment Variables

Variable	Effect
OLLAMA_MODELS	Override Ollama models directory

---

Typical Workflows

Air-gapped transfer

# Machine with internet:
ollama-dl pull llama3:8b -o ./llama3
ollama-dl pack llama3:8b -o llama3-8b.tar.zst

# Transfer llama3-8b.tar.zst to air-gapped machine, then:
ollama-dl unpack llama3-8b.tar.zst --name llama3:8b
ollama run llama3:8b

Verify and repair a suspect install

ollama-dl verify --fix llama3:8b
ollama pull llama3:8b    # re-pull only the corrupt blobs

Audit total disk usage

ollama-dl verify          # lists all models + blob sizes

Compare before upgrading

ollama-dl diff llama3:8b llama3:8b-instruct-fp16
# Decide if the extra download is worth it