Do not require OpenRouter API key when running local models (Ollama, LM Studio, custom URL)

I love the idea here for Claudish. Having used `cline` and feel like I'm going backwards in tooling, being able to use Claude-code with local models feels like a game-changer. However.... ### Problem Claudish requires an OpenRouter API key even when running with a local model such as `ollama/llama3`, LM Studio, or a custom HTTP URL. This blocks users who only want to run local models and do not require any cloud/OpenRouter features, and creates confusion for users who have no need for an OpenRouter account or API key. ``` # tried with `ollama/model` syntax ➜ ~ claudish --model ollama/devstral-small-2:24b OpenRouter API Key Required Get your free API key from: https://openrouter.ai/keys ? Enter your OpenRouter API key: > API key cannot be empty # Also tried with endpoint to ollama ➜ ~ claudish --model http://localhost:11434/devstral-small-2:24b OpenRouter API Key Required Get your free API key from: https://openrouter.ai/keys ? Enter your OpenRouter API key: ``` ### Proposal - Skip OpenRouter API key prompting and key checks when the selected model source is a local provider (e.g., model IDs starting with `ollama/`, `lmstudio/`, or a local HTTP URL). - Only require the OpenRouter API key when the chosen model is actually provided by OpenRouter (e.g., `x-ai/grok-*`, `openai/gpt-*`, etc). - Update CLI startup logic and documentation to clarify when the key is and is not required. - Ideally: fallback to degraded behavior or warnings if OpenRouter-only features are attempted with no key, rather than full CLI exit. ### Benefits - Makes Claudish friendlier for fully local/offline workflows. - Avoids unnecessary prompts for users who will never use OpenRouter cloud models. - Clearer separation of local vs cloud/backend authentication requirements. ### Background Related discussion and code: - [src/index.ts](https://github.com/MadAppGang/claudish/blob/main/src/index.ts) - unconditional API key prompt in interactive mode. - [src/cli.ts](https://github.com/MadAppGa

GitHubDec 19, 2025ToolDevSource

FL score

out of 100

Verdict

VALIDATE

high confidence

Competition

competitors found, emerging market, funded players

Trend

No signal yet

Improve Claudish to allow local LLM usage (Ollama, LM Studio) without requiring an OpenRouter API key, targeting developers prioritizing privacy and cost.

The pain

Developers wanting to use Claudish with local LLMs are frustrated and blocked by the requirement to provide an OpenRouter API key, even when they have no intention of using cloud models, leading to unnecessary friction, privacy concerns, and abandonment of the tool.

The gap

Claudish currently falls behind competitors like Aider, OpenCode, and dedicated local platforms (Ollama, LM Studio) by forcing a cloud API key for local workflows, failing to meet the demand for truly independent, cost-free local LLM experiences.

Build angle

Implement conditional API key prompting and validation logic in Claudish: skip OpenRouter key checks when a local model source (e.g., 'ollama/', 'lmstudio/', local HTTP URL) is specified, only enforcing the key for actual OpenRouter cloud models.

Strengths

Very clear and specific problem for a well-defined niche.
Directly addresses major user frustrations (privacy, cost, unnecessary friction).
Aligns with the growing trend of local-first LLM adoption.

Questions about this idea?

FlyBot reads the scoring and gives you a second opinion on “Do not require OpenRouter API key when running local models (Ollama, LM Studio, custom URL)”.

Open FlyBot

Fly Labs Method

Is the pain real, is there a gap, is it the right time, can one person build it.

Problem clarity: 26
Solution gap: 19
Willingness to pay: 14
Buildability: 8

This idea addresses a clear pain point for local-first LLM users by removing an unnecessary API key requirement in Claudish. While the problem and solution are specific and impactful for user experience, the build complexity within an existing codebase, and the lack of direct monetization or urgency for the problem itself, temper the overall score.

Value Equation

Dream outcome and how likely it feels, against the time and effort it costs.

Addresses a clear market pain in a growing niche but lacks strong differentiation and direct monetization, making it more of an enabling feature than a standalone money-maker.

One-Person Business

Curiosity pull, identity fit, and a path from free value to paid for a solo creator.

A very clear problem for a specific niche, but building it for an existing project and the lack of direct monetization make it less ideal for a solo builder seeking their own sustainable product.

Viral Frameworks

Hook strength, shareability, and how cheaply it can be tested.

Strong problem and value proposition for a clear niche, but lacks a direct business model and relies on contributing to an existing open-source project for distribution and acceptance.

Builder Lens

Evidence the problem exists, timing, defensibility, and a model that fits on a napkin.

Addresses a very specific, real pain point for a clearly defined user, with a clear smallest solution, making the tool more future-fit and competitive.

Why this verdict

The problem is real and articulated well by various sources and the specific prompt example. The demand for local-first LLM is growing due to privacy and cost. However, this is a feature fix for an existing open-source project, not a new product idea with a direct business model for a solo builder. While it improves the tool's competitiveness, the 'buildability' score is lower for an *external* solo builder due to codebase complexity, and the 'willingness to pay' is indirect. Therefore, 'VALIDATE_FIRST' is appropriate to confirm maintainer interest and feasibility within the existing project structure before committing significant time to a non-monetizable contribution or a full fork.

Five lenses, one composite. How scoring works

The angle

This weekend

Submit this as a detailed issue or pull request to the Claudish GitHub repository to gauge maintainer interest and estimated effort for integration.
If maintainer interest is low, consider forking Claudish to implement the feature and market the 'local-first' fork directly.
Engage with Claudish users on Reddit or GitHub discussions to confirm the severity and widespread nature of this blocker.
Estimate the actual effort required to implement this change within the Claudish codebase.

Who is already there, emerging market

OllamaVisit
A lightweight platform that allows users to run large language models locally on their own computer with a single command.
Pricing: Free (open-source), users only pay for tokens if routing to cloud models.
LM StudioVisit
Provides a polished graphical user interface for managing and running local LLMs, making it accessible for non-technical users.
Pricing: Free (for personal use, likely open-source based on user discussions).
ClineFundedVisit
An open-source, agentic CLI coding assistant and VS Code extension designed for local-first workflows and model flexibility.
Pricing: Free (open-source), users only pay for models they run (local or cloud APIs).
AiderVisit
A terminal-native coding agent with excellent Git integration, good at multi-file edits, and supports any model via its OpenAI-compatible API support.
Pricing: Free (open-source), users only pay for tokens (Bring Your Own Key - BYOK). [1]
CursorFundedVisit
An AI code editor (VS Code fork) offering multi-model support, visual diffs, and autocomplete.
Pricing: $20/month. [1]
Continue.devVisit
An open-source IDE extension that works with any LLM provider and supports VS Code and JetBrains, offering custom AI agents and workflow automation.
Pricing: Free (open-source). [1]
OpenCodeVisit
A model-flexible terminal agent, offering features similar to Claude Code among open-source alternatives, supporting 75+ providers including local Ollama. [1, 15]
Pricing: Free (BYOK). [1]
guIDEVisit
A local LLM IDE that works completely offline, allowing users to chat with a local LLM about their code and get inline completions and explanations.
Pricing: Free, no login required. [20]

Funded players

What they charge

The pricing landscape for local LLM development tools is bifurcated between free, open-source solutions and subscription-based cloud alternatives. Open-source tools like Ollama, LM Studio, Cline, Aider, and Continue.dev are generally free, with users only incurring costs for the models themselves (either local hardware investment or cloud API usage if integrated). [1, 22, 24] Cloud-based coding assistants, such as Cursor and GitHub Copilot, typically charge monthly subscriptions ranging from $10 to $20 per month. [1] Some on-premise AI hosting solutions for local LLMs offer plans starting from $500/month for small teams. [29] The overall trend leans towards free tools for local inference, with many users seeking to avoid recurring API costs. [5, 18]

Do not require OpenRouter API key when running local models (Ollama, LM Studio, custom URL)

Excessive CAPTCHA challenges degrading GitHub login UX for enterprise users

GTA modders tediously manually edit assembly files for modpacks

BreezePDF – Free, in-browser PDF editor