Connecting an AI provider
Updated Oct 11, 2026 · 3 min read
AI tagging runs on a provider you choose: your own OpenRouter, OpenAI, Anthropic or Google Gemini key, Ollama on this computer, or your ChatGPT plan via Codex.
Meshory's AI descriptions, tags and semantic search are opt-in, and there's no Meshory AI subscription. You choose who does the work in Settings → AI tagging, which has two groups, Descriptions and tags and Semantic search, each with its own provider. A provider you reach with an API key bills you directly, per use; Ollama runs on your own computer; and the Codex CLI runs on your ChatGPT plan.
Which provider to pick
- OpenRouter: one key reaches models from every major provider. It's the default, and in Meshory 1.11.0 the only provider semantic search works with.
- OpenAI API, Anthropic API or Google Gemini API: a key from that company's developer console, billed by them per use. API billing is separate from a ChatGPT or Claude app subscription.
- Ollama: runs a model on this computer, with no key and no bill. Meshory connects to the Ollama server on this computer at the port you set, and you install the model in Ollama yourself; Meshory never downloads one.
- Codex CLI: uses your ChatGPT subscription through OpenAI's Codex command-line tool, installed on this computer and signed in with ChatGPT. Each run counts against your plan's Codex usage limits, and Meshory never sees your sign-in.
Semantic search uses OpenRouter in Meshory 1.11.0; the other providers in its list show Search calibration pending. With only Ollama or the Codex CLI set up, descriptions and tags work but semantic search doesn't.
Get a key from OpenRouter
- 1Sign up at openrouter.ai. One account gives you access to models from every major provider.
- 2Add a few dollars of credit on the Credits page. OpenRouter is prepaid, so you can never spend more than you've loaded.
- 3Create a key at openrouter.ai/keys, name it something like “Meshory”, and copy it. It starts with sk-or-. You can give the key its own spending limit as an extra guardrail.
New OpenRouter accounts include a very small free allowance, and some models have free rate-limited variants, enough for a first test run, though a real library will want a few dollars of credit.
Add a key to Meshory
- 1Open Settings → AI tagging and, under Descriptions and tags, pick the provider.
- 2Paste the key into its API key field and click Save. It's stored encrypted via your system keychain, and the field shows Key saved instead of the key.
- 3Pick a Generation model, then click Test descriptions and tags. Generation connection OK means you're set.
- 4For semantic search, choose OpenRouter under Semantic search. It uses the key you saved above; click Test semantic search to check it.
Ollama on this computer
Install Ollama and a model in it first: Meshory lists the models already installed and never pulls one. Choose Ollama as the provider, change the port if you moved Ollama off its default of 11434, click Refresh installed models, pick one and test it. Leave that model alone while Meshory is working: if it changes, work stops until you test again.
Codex CLI with your ChatGPT plan
Install the Codex CLI with npm install -g @openai/codex, run codex login in a terminal and sign in with ChatGPT, then choose Codex CLI as the provider. Meshory finds the Codex installed on this computer, and says what's missing when it can't run it. Generation model starts on Codex default, which lets Codex choose; larger models use more of your plan per run.
Running it
From there, run tagging on your whole library, a folder, or a selection, whenever you want. The AI menu in the library toolbar shows the provider and model, what processing new files will cost (in dollars, free on this computer, or in Codex runs) and a log of requests with their live status. If a run fails immediately, test the connection again and make sure a paid account has credit.
Nothing AI-related happens until you set up a provider and start it. Until then, your library never talks to a model provider.