🦙 tallow-ollama-cloud-extension

Connect Tallow to Ollama Cloud with live model discovery

A Tallow extension that registers Ollama Cloud as a first-class provider. Automatically discovers all available cloud models via the OpenAI-compatible /v1/models endpoint — no manual models.json editing required.

Why use this? Instead of manually maintaining a static list of Ollama Cloud models in ~/.tallow/agent/models.json, this extension ships 10 built-in defaults and adds /login ollama-cloud and /ollama-refresh for on-demand model discovery. No polling, no background timers — tallow stays lightweight.

After a successful /login ollama-cloud, the model list is automatically refreshed with the new API key.

For the Tallow project, see: github.com/dungle-scrubs/tallow

Official documentation:

• Extensions Overview
• Creating Extensions

Table of Contents

• Quick Start
• How Extensions Load
• Installation
• What You Get
• Commands
• Configuration
• FAQ
• Troubleshooting
• Contributing
• License

Quick Start

# 1. Install the extension into ~/.tallow/extensions/
git clone https://github.com/Hackbard/tallow-ollama-cloud-extension.git
cp -r tallow-ollama-cloud-extension/extensions/ollama-cloud ~/.tallow/extensions/

# 2. Start tallow
tallow

# 3. Log in with your API key (do this first!)
/login ollama-cloud
# → Paste your key from https://ollama.com/settings/keys

# 4. Pick an Ollama Cloud model
/model
# → Select any ollama-cloud/... model

That's it. The model list stays in sync with Ollama Cloud automatically.

How Extensions Load

Tallow discovers extensions from three locations in order (docs):

1. Bundled — shipped in tallow's extensions/ directory (loaded automatically)
2. User global — ~/.tallow/extensions/ (available in all sessions)
3. Project local — .tallow/extensions/ (scoped to a single project)

If a user extension has the same name as a bundled one, the user version overrides it.

Inspect installed extensions

tallow extensions              # table view of all extensions
tallow extensions --json       # machine-readable catalog
tallow extensions ollama-cloud # detail view for this extension

Installation

One-liner (Recommended)

curl -fsSL https://raw.githubusercontent.com/Hackbard/tallow-ollama-cloud-extension/main/install.sh | bash

Or install into a specific project:

curl -fsSL https://raw.githubusercontent.com/Hackbard/tallow-ollama-cloud-extension/main/install.sh | bash -s /path/to/project

npm

npm install -g tallow-ollama-cloud-extension
tallow-ollama-cloud-extension

Or project-local:

npm install tallow-ollama-cloud-extension
npx tallow-ollama-cloud-extension .

Manual

git clone https://github.com/Hackbard/tallow-ollama-cloud-extension.git
./tallow-ollama-cloud-extension/install.sh

What You Get

| Capability | Description | |------------|-------------| | Live model discovery | /ollama-refresh fetches all available Ollama Cloud models from https://ollama.com/v1/models on demand | | Login-triggered refresh | After /login ollama-cloud succeeds, models are automatically refreshed with the new key | | /login ollama-cloud | Interactive OAuth login prompt. Stores your API key securely in ~/.tallow/agent/auth.json | | /ollama-refresh | Manual refresh command when you want the latest models | | Smart defaults | Ships 10 built-in models — works immediately with zero network calls | | Lightweight | No auto-refresh, no background polling, no timers — tallow stays fast |

Smart model heuristics

The extension guesses capabilities for every discovered model:

| Model ID pattern | Detected capability | |------------------|---------------------| | *-vl-*, *vision*, *llava*, gemini* | input: ["text", "image"] | | *r1*, *thinking*, *cogito*, deepseek-v4* | reasoning: true | | kimi-k2* | contextWindow: 256_000 | | everything else | contextWindow: 128_000, input: ["text"] |

These are best-effort heuristics. If a model is mis-tagged, use /model details or open an issue.

Commands

| Command | Description | |---------|-------------| | /login ollama-cloud | Prompt for your Ollama Cloud API key, store it, and auto-refresh models | | /ollama-refresh | Manually fetch the latest model list from Ollama Cloud |

Configuration

Environment variable

You can skip /login entirely by setting your key as an environment variable:

export OLLAMA_CLOUD_API_KEY="sk-..."
tallow

When this variable is present, the extension uses it directly and still refreshes the model list automatically.

Per-project override

If you need to pin a specific Ollama Cloud model in a project, add this to .tallow/settings.json:

{
  "defaultProvider": "ollama-cloud",
  "defaultModel": "qwen3.5:397b"
}

FAQ

Q: Is the model list fetched with or without authentication?
A: The model registry at https://ollama.com/v1/models is public — no API key is needed to discover models. You only need a key for the actual chat completions (POST /v1/chat/completions).

Q: Where is my API key stored?
A: It is stored in ~/.tallow/agent/auth.json by Tallow's built-in OAuth credential manager. The extension never writes keys to its own files.

Q: Do I need to run /ollama-refresh manually?
A: No. The extension refreshes automatically every 5 minutes and on every session_start. The command exists only if you want an immediate refresh (e.g., after a new model announcement).

Q: What happens if Ollama Cloud is down?
A: The extension falls back to a built-in list of 10 known models. Your session keeps working.

Q: Can I use this with local Ollama instead?
A: No — this extension is specifically for Ollama Cloud (ollama.com). For local Ollama (localhost:11434), add a static provider block to ~/.tallow/agent/models.json instead:

{
  "providers": {
    "ollama": {
      "baseUrl": "http://localhost:11434/v1",
      "api": "openai-completions",
      "apiKey": "ollama",
      "models": [{ "id": "llama3.3" }]
    }
  }
}

Q: A new model is not showing the right capabilities. How do I fix it?
A: Open an issue with the model ID. The heuristics are regex-based and may need an update for new naming conventions.

Troubleshooting

"No API key configured" when trying to chat

You have selected an Ollama Cloud model but have not authenticated:

/login ollama-cloud

Or set the environment variable before starting:

export OLLAMA_CLOUD_API_KEY="sk-..."

Model list is empty or stale

Run a manual refresh:

/ollama-refresh

If that fails, check your network connection to https://ollama.com/v1/models. The extension uses a 8-second timeout — slow networks may need a retry.

Extension does not appear after install

Ensure you ran /reload or restarted Tallow. Extensions are discovered at startup.

Contributing

PRs welcome! Before opening a PR:

1. Test the extension locally by dropping it into ~/.tallow/extensions/ollama-cloud/.
2. Run /reload and verify /model lists Ollama Cloud models.
3. Check that /login ollama-cloud stores credentials and that chat completions work.

If you find a model with incorrect capability detection, include the exact model ID in your issue.

License

MIT