OpenRouter Free Agents Skill

Updated September 27, 2026.

This GitHub Copilot skill manages free OpenRouter coding models in VS Code Copilot Chat. The skill scans and synchronizes available free OpenRouter models and populates the model selector with a curated list of currently free models for quick access in VS Code Copilot Chat under Other Models → Free OpenRouter. The skill updates the custom provider, pins active free models, unpins expired models, and removes legacy generated agent files to keep the agents menu clean.

imageFree OpenRouter models in the Copilot Chat model selector

The OpenRouter Free Agents Skill optionally evaluates the models for health and coding capability, providing a summarized chat report and offers a durable Markdown report for future reference.

Install

For GitHub Copilot, choose an installation method below.

Install as a plugin

In VS Code, run Chat: Install Plugin From Source from the Command Palette and enter:

https://github.com/MindsEyeProducts/OpenRouter-Free-Agents-SKILL

Or use the Copilot CLI to add the marketplace and install the plugin:

copilot plugin marketplace add MindsEyeProducts/OpenRouter-Free-Agents-SKILL
copilot plugin install openrouter-free-agents@mindseye-skills

The plugin includes the same /openrouter-free-agents skill described below. Complete the prerequisites, including OpenRouter setup, before running it. To update a CLI installation later, run copilot plugin update openrouter-free-agents.

Install as a standalone skill

Copy only skills/openrouter-free-agents/ into your personal skills directory. That folder contains the complete skill, script, and references and works without the rest of this repository. Install either the plugin or the standalone skill to avoid duplicate commands.

Clone with Git

git clone https://github.com/MindsEyeProducts/OpenRouter-Free-Agents-SKILL.git
New-Item -ItemType Directory -Force "$HOME/.copilot/skills" | Out-Null
Copy-Item -Recurse ./OpenRouter-Free-Agents-SKILL/skills/openrouter-free-agents "$HOME/.copilot/skills/"

Download ZIP

1. Download the latest ZIP, or select Code → Download ZIP on GitHub.
2. Extract the ZIP and open OpenRouter-Free-Agents-SKILL-main/skills/.
3. Copy its openrouter-free-agents folder into %USERPROFILE%\.copilot\skills\ on Windows, creating the destination skills directory if needed.

The resulting file path should be %USERPROFILE%\.copilot\skills\openrouter-free-agents\SKILL.md, with scripts and references alongside it.

Updating an older standalone installation

Version 1.0.2 moves the skill from the repository root into skills/openrouter-free-agents/ so plugin intake discovers it correctly. If you previously cloned the repository directly into your personal skills directory, move that old installation to a backup location outside any skills directory, then install the nested folder using the steps above. A git pull alone no longer leaves a root SKILL.md in that older installation layout. Keep a source checkout outside the personal skills directory and copy the nested skill folder when updating.

Plugin users can continue using copilot plugin update openrouter-free-agents. The repository command python -B scripts/sync_openrouter_agents.py remains available as a compatibility launcher.

Run

1. Run the skill in Copilot Chat

Open your project in VS Code, open Copilot Chat, and enter:

/openrouter-free-agents

This synchronizes the free model list and offers an optional coding capability and health evaluation afterward.

To request synchronization, evaluation, and a saved report together, you can also say:

Sync the free OpenRouter models, evaluate their coding capabilities and health, and save the Markdown report in this project.

Or use the combined command:

/openrouter-free-agents --eval-md

With the combined command, evaluation and saving are already requested, so the agent skips the later Y/N questions.

2. Review and allow the script

Review each proposed command before selecting Allow. In the captured walkthrough, the agent first requested permission to check the script path:

imageCopilot approval prompt for checking the sync script path

It then requested permission to run the synchronization script:

imageCopilot approval prompt for running the model synchronization script

These screenshots and the example results below were captured on September 26, 2026. The number of approval prompts depends on your VS Code settings and how the skill is invoked. Model availability, counts, and health values will change between runs.

3. Review the synchronization results

The agent reports the configured models, newly pinned and retained models, models that were unpinned, and any legacy agent files removed. It also lists the active free models and their tool support and context sizes.

Example synchronization summary and active model list — September 26, 2026

Free OpenRouter Models Sync Summary

The synchronization script has queried OpenRouter and updated VS Code:

• Configured Models: 21 active free models configured under the Other Models -> Free OpenRouter group in chatLanguageModels.json.
• Model Selector Pinning (state.vscdb):
  • Newly Pinned: 18 free models
  • Retained: 3 existing free models
  • Unpinned / Cleaned: 4 models (no longer free or expired)
  • Total Pinned Models: 24
• Custom Agents Status: Clean. Custom Agents are not populated in GitHub Copilot Set Agents, and no legacy .agent.md files were found.

Active Free Models

4. Optionally evaluate coding capability and health

After a bare /openrouter-free-agents invocation, the agent asks:

Would you like a coding capability and health evaluation of these free models? (Y/N)

Type Y to view the evaluation, or N to finish without it. The evaluation combines live OpenRouter health data with maintained coding tiers and model notes.

If prompted, review the evaluation command and press Allow:

imageCopilot approval prompt for the evaluation preview command

The agent presents model tiers, technical specifications, suggested use cases, and a health table. The following is the example evaluation from the How To walkthrough.

Example coding capability and health evaluation — September 26, 2026

OpenRouter Free Models: Coding Capability & Health Evaluation

Generated: September 26, 2026
Scope: Models in VS Code Copilot Chat (Other Models -> Free OpenRouter)
Total Evaluated Models: 21

1. Overview & Evaluation Criteria

This document provides a technical evaluation of the active free models available on OpenRouter, assessed for real-world software engineering, code generation, refactoring, agentic tool use (reading/editing files, executing commands, running tests), and real-time operational health.

Evaluation Dimensions

• Coding Benchmarks & Training: Dedicated code fine-tuning vs. general-purpose instruct.
• Operational Health: Real-time provider availability, 5m and 24h uptime %, and response latency.
• Tool Calling (tools): Critical for GitHub Copilot Chat tool use (search, read, edit, terminal execution).
• Reasoning Architecture: MoE parameter efficiency, dedicated reasoning tokens, chain-of-thought support.
• Context Window: Ability to ingest entire multi-file codebases, logs, and documentation.
• Latency & Throughput: Suitability for fast iterative programming vs. deep architectural design.

2. Tier Summary & Technical Specifications

Health Legend: 🟢 Healthy / High Availability (≥95% uptime)  |  🟡 Degraded / Reduced Availability (<95% uptime)  |  🔴 Offline / Outage

3. Recommended Models by Use Case

1. Daily Coding, Bug Fixing & Refactoring

• Primary Pick: poolside/laguna-s-2.1:free (Poolside: Laguna S 2.1)
  • Built specifically for coding agents, diff generation, and terminal tool calls.
  • Strong benchmark results on Terminal-Bench 2.1 (70.2%).
• Secondary Pick: cohere/north-mini-code:free (Cohere: North Mini Code)
  • Cohere’s specialized agentic coding model with strict syntactic precision and clean documentation generation.

2. Complex Architecture, System Design & Large Repositories

• Primary Pick: thinkingmachines/inkling:free (Thinking Machines: Inkling)
  • 41B active parameters / 975B total with a 1,048,576-token context window.
  • Excels at tracing complicated bugs across large workspaces and designing modular architectures.
• Alternative: nvidia/nemotron-3-ultra-550b-a55b:free (NVIDIA: Nemotron 3 Ultra)
  • 55B active parameters; hybrid Transformer-Mamba architecture capable of high-throughput reasoning.

3. Fast Interactive Iteration (Low Latency)

• Primary Pick: poolside/laguna-xs-2.1:free (Poolside: Laguna XS 2.1)
  • Activates only 3B parameters per token for near-instant responses on unit test generation and single-function refactoring.
• Alternative: google/gemma-4-26b-a4b-it:free (Google: Gemma 4 26B A4B)
  • 3.8B active parameters; balanced speed and reasoning for quick shell scripts and boilerplate.

4. Multimodal UI & Visual Design Work

• Primary Pick: google/gemma-4-31b-it:free (Google: Gemma 4 31B) or minimax/minimax-m3:free (MiniMax M3)
  • Native image understanding enables inspecting screenshots, wireframes, and design specs to produce matching front-end code (HTML/Tailwind, React, Vue, Flutter).

4. How to Select Models in VS Code Copilot Chat

1. Open GitHub Copilot Chat in VS Code (Ctrl+Alt+I or click the Chat icon in the Activity Bar).
2. Click the Model Selector dropdown at the bottom of the chat panel.
3. Select your model under Other Models -> Free OpenRouter (or under Pinned Models).
4. Recommended starting model: Poolside: Laguna S 2.1 (free) or Cohere: North Mini Code (free).

5. Real-Time Model Health & Operational Status

Telemetry pulled directly from OpenRouter endpoint monitoring:

5. Save the evaluation

After showing an optional evaluation, the agent asks:

Would you like to save this evaluation as OpenRouter_Free_Models_Coding_Capability.md in the current workspace? (Y/N)

Type Y to save a permanent Markdown copy in your project's root directory, or N to keep only the chat preview. The agent saves the displayed evaluation without fetching health or synchronizing again.

When you use --eval-md upfront, the report is saved as part of the combined run. The agent confirms the location and summarizes the saved evaluation.

Example summary of the saved evaluation — September 26, 2026

• Tier S Models: cohere/north-mini-code:free, poolside/laguna-s-2.1:free, and thinkingmachines/inkling:free (all 🟢 Online).
• Tier A Models: google/gemma-4-31b-it:free, nvidia/nemotron-3-ultra-550b-a55b:free, and poolside/laguna-xs-2.1:free (all 🟢 Online).
• Operational Health: Detailed 5-minute and 24-hour uptime metrics, response latency tiers, and upstream provider status for all 21 active free models.

See OpenRouter_Free_Models_Coding_Capability.md for an example saved report.

6. Reload VS Code and select a model

After synchronization and any chosen evaluation or save steps finish, press Ctrl+Shift+P (or F1) and run Developer: Reload Window:

imageVS Code Command Palette with Developer: Reload Window selected

Then open the model selector at the bottom of Copilot Chat and choose a model under Other Models → Free OpenRouter:

imageFree OpenRouter models displayed in the Copilot Chat model selector

If you find this skill useful, consider buying me a ☕.

Other workflows

The staged workflow can require separate approvals for synchronization, evaluation, and saving. Use --eval-md upfront to complete synchronization and report generation in one script invocation.

Direct terminal alternative

With the personal Copilot skill installed, run this from your project directory:

python -B "$HOME/.copilot/skills/openrouter-free-agents/scripts/sync_openrouter_agents.py" --eval-md

If running from another directory, add --workspace-dir "C:\path\to\your\project". An optional filename can follow --eval-md; for example, --eval-md "free-model-report.md" saves that filename in the selected project.

--eval-md performs synchronization as well as report generation. --evaluate previews the report without modifying VS Code or saving a file. --no-db-sync skips only database updates; provider configuration and legacy file cleanup still run.

Prerequisites

Before installing and running this skill, make sure you have:

• VS Code with Copilot Chat available. Use a desktop installation with support for agent skills, terminal commands, and custom model providers. The setup instructions in this README use Windows and PowerShell.
• Python 3 available in your terminal. Confirm that python --version reports a Python 3 installation. The bundled script uses the Python standard library; no additional Python packages are required.
• An OpenRouter account and API key configured in VS Code. Create a key in your OpenRouter account settings, then add OpenRouter through Chat: Manage Language Models. Follow the OpenRouter setup guide for Copilot.
• Internet access to OpenRouter. The skill retrieves the model catalog and health data from https://openrouter.ai/api/v1.
• A local project folder open in VS Code. Run the skill from that project so any requested Markdown report is saved in the intended location. The synchronization script also needs permission to update your VS Code user configuration and model pinning data.

Configure OpenRouter before the first sync

Open the Command Palette with Ctrl+Shift+P, run Chat: Manage Language Models, and add the OpenRouter provider with your API key. Confirm that an OpenRouter model is available in the model selector before running this skill. See VS Code's model configuration documentation.

The current script expects an existing chatLanguageModels.json configuration and reuses a configured provider's API-key reference. It does not create your OpenRouter account or register a new API key. On Windows, it looks for the standard VS Code user configuration under %APPDATA%\Code\User\.

Optional: Git

Install Git if you plan to use the Clone with Git installation method. The Download ZIP

Contents

• skills/openrouter-free-agents/SKILL.md — agent instructions and supported workflows
• README.md — installation and usage walkthrough with screenshots, example results, and commands
• skills/openrouter-free-agents/scripts/sync_openrouter_agents.py — model synchronization, evaluation preview, and Markdown report generation
• scripts/sync_openrouter_agents.py — compatibility launcher for the previous repository command
• assets/run/ — screenshots used in the usage walkthrough
• skills/openrouter-free-agents/references/openrouter-api.md — OpenRouter API notes
• skills/openrouter-free-agents/references/agent-pinning.md — VS Code provider and model pinning details
• OpenRouter_Free_Models_Coding_Capability.md — generated coding capability and health report
• tests/test_sync_workflows.py — checks for synchronization, preview, reporting, and dry-run behavior

Development checks

Python 3 is sufficient to run the skill and its workflow tests:

python -B -m unittest discover -s tests -v

For plugin packaging checks, use Node.js 24 and npm 11.11.1 or newer:

npm ci --ignore-scripts
npm test

These checks use the same Vally 0.12.0 runLint API as the Awesome Copilot intake. They lint a complete checkout staged under the name submission and a standalone copy of the skill, require exactly one discovered skill, and check marketplace version consistency. CI runs both suites on Windows and Linux. Node.js is needed only for development validation.

Support

If you find this skill useful, consider buying me a ☕.

License

MIT