Skip to main content

Custom API Setup

Filo Plus includes Filo-managed AI. If you use the Free plan instead, you can connect your own AI provider or local model in Filo Desktop and continue using Filo AI features with your own API quota.

Custom API is configured per device in Filo Desktop. You do not need Filo Plus to use your configured provider, but you do need a supported model and any balance or billing required by that provider. Provider charges are separate from Filo.

Choose how you want to use Filo AI

Filo Plus

The simplest option. Filo manages the models, service, and quota, so there is no external API account, key, or provider bill to maintain.

Custom API on the Free plan

Connect a supported provider, OpenAI-compatible endpoint, or local model. Filo supplies the AI experience; your provider supplies the model and bills the usage.

AI features available with Custom API

Summaries and automatic to-dos

Summarize emails and threads, identify action items, and automatically turn relevant messages into to-dos.

AI conversation

Ask questions, explore email context, and use the Filo AI conversation experience with your selected model.

Compose and quick reply

Draft new messages, generate replies, rewrite text, and adapt tone or wording before you send.

Translation and other AI actions

Use AI translation and the other provider-backed AI tools included in your installed Filo Desktop version.

Before you start

  • •Update Filo Desktop to the latest version. If Settings does not show API Configuration or Custom AI (BYOK), your installed build does not include the feature yet.
  • •For a cloud provider, create an account, enable billing or add balance when required, and create an API key. A ChatGPT, Claude, or Gemini consumer subscription does not automatically include API usage.
  • •Choose a model that supports the work you need. AI conversation and agent actions work best with models that support tool calling and a sufficient context window.
  • •For a local provider, install it, download a model, and start its local server before testing the connection in Filo.

Supported provider types

Direct cloud providers

OpenAI, Anthropic, Google Gemini, DeepSeek, xAI, Qwen, Mistral, Z.AI, Doubao, Moonshot, MiniMax, Together, Fireworks, Groq, Cerebras, SambaNova, Nebius, and Hugging Face.

Routers and gateways

OpenRouter, Requesty, Vercel AI Gateway, AIHubMix, and compatible routing services let you access one or more model families through a single account.

OpenAI-Compatible

Use a custom OpenAI-compatible service by entering its API Base URL, API key, and exact model ID. Both a base URL such as https://api.example.com/v1 and a full /chat/completions URL are accepted.

Local and self-hosted

Ollama and LM Studio run on this computer without an API key. LiteLLM can run locally or remotely; its API key is optional, while remote deployments must use HTTPS.

What each field means

Provider

Choose the company or endpoint that will receive Filo AI requests. Use OpenAI-Compatible for a custom compatible endpoint.

API Base URL

The root address of the provider API. Keep the official default unless your provider documents another URL. Custom and self-hosted providers require the exact address.

API key

The secret credential from your provider. It is not your normal account password. Ollama and LM Studio do not need one; LiteLLM may make it optional.

Model

The model Filo will call. Select a listed model or enter the exact model ID, including any provider namespace such as provider/model when required.

Reasoning effort

When shown, Low is faster and usually cheaper, while High spends more time and tokens on difficult tasks. Medium is a good default.

Set up Custom API step by step

  1. Open Filo Desktop and go to Settings → API Configuration.
  2. Find and select your provider. Choose OpenAI-Compatible for a custom compatible service, or Ollama, LM Studio, or LiteLLM for a local or self-hosted setup.
  3. Check the API Base URL. Official providers are prefilled. For OpenAI-Compatible, enter the provider base URL or full /chat/completions URL. Remote URLs must use HTTPS; local Ollama and LM Studio addresses must point to localhost or another loopback address.
  4. Paste the API key from your provider. Skip this for Ollama or LM Studio; add it for LiteLLM only when your deployment requires one.
  5. Select a model. If no catalog appears, enter the exact model ID from the provider or local server.
  6. Choose Reasoning effort when that option is available. Leave it at Medium if you are unsure.
  7. Select Test connection. Filo sends a small verification request and saves the credential only after the provider accepts the configuration.
  8. After the provider shows Ready, turn on Enable Custom AI.
  9. Open an email and try a summary, to-do extraction, AI conversation, or draft. If you change the key, URL, or model later, test the connection again.

What changes when it is enabled

  • •Supported Filo Desktop AI features use the selected provider and model, including summaries, automatic to-dos, AI conversation, compose and reply tools, and translation.
  • •The setting and credential stay on this device and do not automatically sync to your other computers or mobile devices.
  • •Turning off Enable Custom AI switches Filo back to Filo-managed AI. Access to that service then follows your Free or Filo Plus entitlement.
  • •Custom API removes the need for a Filo AI quota on provider-backed requests, but it is not unlimited: your provider balance, rate limits, model limits, and regional availability still apply.

Security and billing

  • •Your credential is stored securely on this device and is sent only to the provider endpoint you select.
  • •Requests may include the email text, thread context, draft, attachment content, or task information needed to complete the AI action.
  • •Do not share screenshots or exported settings that contain your API key or authorization token.
  • •Your provider controls pricing, retention, training, regional processing, rate limits, and model availability. Review its data policy before sending sensitive mail.
  • •Use a dedicated key where possible, set a spending limit in the provider console, and rotate the key immediately if it may have been exposed.

Troubleshooting

  • •401 or 403 means the key was rejected. Check that the key is active, belongs to the selected provider, and has permission to use the model.
  • •429 means the provider rate limit or balance was reached. Wait, add credit, raise the provider limit, or choose another model.
  • •404 usually means the Base URL or model ID is wrong. Confirm both against the provider documentation and avoid adding the same path twice.
  • •A blocked-address or HTTPS error means the endpoint does not meet Filo network-safety rules. Remote services must use HTTPS; Ollama and LM Studio must use a loopback address.
  • •If a local connection fails, start the provider server, confirm its port, and make sure the selected model has finished downloading.
  • •If basic prompts work but an agent action fails, choose a model with tool-calling support and a larger context window, then test again.
  • •If API Configuration is missing, update Filo Desktop and check whether the feature is included in your build.