Wire Logseq's AI plugins to any catalog model.

Quick answer: three settings in the plugin config.

Install an AI plugin from Logseq's marketplace, open its settings, and fill the standard trio: the endpoint or base URL field with https://api.apisrouter.com/v1, the API key field with your APIsRouter key, and the model field with an exact catalog id such as claude-haiku-4-5-20251001. Two concrete examples from the ecosystem. The long-standing gpt3-openai plugin (briansunter/logseq-plugin-gpt3-openai) exposes a chatCompletionEndpoint setting, defaulting to OpenAI's host, plus a free-text model setting, so both redirect cleanly to a gateway. The newer logseq-copilot plugin was built for OpenAI-compatible APIs from the start, with API Endpoint, API Key, and Model settings and a case-sensitive model field. Field names differ per plugin; the pattern does not.

API / Chat Completion Endpoint:
  https://api.apisrouter.com/v1

API Key:   sk-YOUR-APISROUTER-KEY

Model:     claude-haiku-4-5-20251001
           (case-sensitive, free text)

How Logseq plugins use the endpoint.

Logseq is the open-source outliner for networked notes, and its AI plugins hook into the editor as slash commands and block actions: summarize this block, continue writing here, answer from the selection, generate a outline from a page. Each action is one chat-completions request carrying the relevant blocks as context, sent to whatever endpoint the plugin's settings declare. An honest note on the ecosystem, because it is smaller than Obsidian's: Logseq AI plugins are community projects with uneven maintenance. The gpt3-openai plugin is the most established but has seen quiet stretches between releases; logseq-copilot is lighter but purpose-built for OpenAI-compatible endpoints. Check a plugin's recent activity before depending on it daily, and expect to read its settings panel rather than a polished setup wizard. The gateway side is the stable half of this pairing. Because the endpoint speaks the standard chat-completions dialect and serves every catalog id through one key, a plugin configured once can switch between a Claude id and a DeepSeek id by editing one settings field, no reinstall, no new account.

Choosing models for an outliner workflow.

Outliner AI usage is bursty rather than constant: nothing for an hour, then a run of slash commands while writing. Prepaid per-token billing fits that shape better than any subscription, and the console's per-model view shows what your actual cadence costs on each id.

• claude-haiku-4-5-20251001 as the everyday driver: block summaries, quick rewrites, and answers grounded in a page of context.
• claude-sonnet-4-6 for heavier synthesis, drafting long sections from an outline, or anything you will publish.
• gpt-5.4-mini as the GPT-family volume pick if your prompts were shaped on OpenAI behavior.
• deepseek-v4-flash for bulk passes, like summarizing every journal page in a month, at the floor rate.
• glm-5 worth testing for bilingual graphs, especially Chinese-English mixed notes.

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

Failure modes specific to Logseq plugins.

The division of labor when debugging: curl proves the endpoint, key, and id in two commands, and everything else is the plugin's configuration or compatibility. That order saves the most time in a plugin ecosystem where the tooling layer is the moving part.

• Command silently does nothing: open the plugin's settings and confirm the endpoint field actually saved; some plugins require toggling the plugin off and on after settings changes.
• Model errors with a not-found message: the model field is case-sensitive in at least one popular plugin, and ids must match the catalog listing exactly. Copy from /v1/models, not from memory.
• Endpoint field expects a different path depth: most plugins want the base ending in /v1 and append routes themselves; if one demands a full path, give it /v1/chat/completions explicitly and note which style your plugin uses.
• 401 errors: the key field holds whitespace or an old key. Curl the models listing with the same key to confirm outside Logseq.
• Plugin broken after a Logseq update: community plugins lag app releases sometimes. Check the plugin's repository issues before assuming the endpoint changed; the curl check proves the gateway half in seconds.

Who routes Logseq through a gateway.

• Daily journalers running summaries and rewrites over their graph on a volume-tier id.
• Writers drafting from outlines who want a flagship id on tap for the passages that matter.
• Bilingual note-takers testing GLM and DeepSeek ids against Claude on mixed-language blocks.
• Privacy-minded users who picked Logseq for local files and want their AI traffic consolidated behind one auditable key.
• People already routing editors or terminals through a gateway who want notes on the same balance and usage log.

Verify the endpoint, then test one block.

Run the two-command check with the exact model id you configured, then trigger one slash command on a small block. If curl passes and the plugin fails, the issue is the plugin's settings or version, and its repository is the right place to look. Once commands flow, the APIsRouter console records model, tokens, and spend per request. Notes workloads are light compared to coding agents, so most Logseq users find the console mostly confirms the bill is small; the value is knowing that from data rather than hoping.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | head -50

curl -s https://api.apisrouter.com/v1/chat/completions \
  -H "Authorization: Bearer $APISROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-haiku-4-5-20251001",
       "messages":[{"role":"user","content":"Summarize: spaced repetition beats cramming."}]}'

FAQ