Usage Statistics, Import/Export and Backup

This page explains the settings pages Usage (token statistics, consumption overview, requests per model) and Import/Export (configuration backup, JSON export, import from LibreChat or LiteLLM), and what you can and cannot recover from a backup.

What this does

• Usage shows how many AI requests and tokens the installation consumed in the current and the previous month, broken down by model — site-wide, without any person-related data. It is a transparency counter, not a billing basis.
• Import/Export downloads the configuration as a JSON file (without API keys) and fills the settings forms from an AnyLLM export, a librechat.yaml or a LiteLLM config.yaml.

Both pages belong to the one configuration shared by Jira and Confluence.

Who can do this

Product administrators (Jira or Confluence administration) on the settings page AnyLLM Chat — Settings — see Opening the Settings and Initial Setup.

Reading the usage statistics

Purpose — keep an eye on consumption per model.

Steps

1. Open Settings → Usage. While loading, the page shows Loading statistics …
2. Read the two cards Current month (YYYY-MM) and Previous month (YYYY-MM) — months are counted in UTC.

What each card shows

• A total line: {requests} requests · {total} tokens ({in} input / {out} output).
• A table Model | Requests | Input tokens | Output tokens, sorted by total tokens (highest first). Models are listed as <connection ID>::<model> — the connection's technical ID, not its display name; the same model under two connections appears twice.
• If more than 20 models were used in a month, the remaining ones are aggregated in a row literally named (weitere) (German for "others", shown in both languages).
• No data yet. appears before the first request of that month has been counted.

What counts as a request

• Every call to the AI service, not every message: a question that makes the AI look things up in Jira/Confluence or on the web counts one request per round (up to 6 per message).
• Requests from every surface: chat, quick actions, chat bar, temporary chats and each ↻ Refresh of a dashboard gadget.
• Not counted: the short extra call that generates a conversation's title.

Prerequisite — the figures come from what the AI service reports back in its streaming responses. OpenAI-compatible services must return the usage object in the stream (the app requests it); Azure OpenAI and Anthropic report it natively. If a service reports nothing, its models stay at zero — ask your provider whether its streaming API returns usage.

Common problems

Exporting the configuration

Purpose — create a backup or move the configuration to another site.

Steps

1. Open Settings → Import/Export.
2. In the card Export, click ↓ Export as JSON.

Expected result — your browser downloads anyllm-export-YYYY-MM-DD.json. It contains the saved connections (name, endpoint, models — no API keys and no provider type), the default model, the AI access policy, the rate limit, the site agents (without their fixed model) and the templates/skills (name and text only — the skill setting and agent assignment are not included).

Importing a configuration

Purpose — restore a backup, move to another site, or take over endpoints, models and agents from LibreChat or LiteLLM.

Supported formats

Steps

1. Open Settings → Import/Export.
2. In the card Import, click Choose file (JSON/YAML) … and pick the file — the page confirms File "{name}" loaded — now check. — or paste the content into the field … or paste content.
3. Click Check. On success the page reports Detected: {format} (AnyLLM-Export, librechat.yaml or LiteLLM config.yaml); warnings from the parser are listed underneath.
4. Click Apply ({conn} connections, {agents} agents, {prompts} prompts). The page reports Applied to the forms — please review in each section, add keys/endpoints and SAVE.
5. Open Connections: review the imported cards (they are appended after your existing ones), set the Provider type, complete the Endpoint, enter each API key, then click Save connections and confirm the Atlassian approval dialog.
6. Open Agents: review the cards, set a Model where wanted, click Save agents.
7. Open Prompts & Skills: review the cards, tick the skill checkbox again where needed, click Save prompts.
8. Open General: set the default model, AI access and rate limit again, click Save.
9. Open the settings in the other product too and approve the endpoint addresses there (Endpoint approvals) — a yellow notice with Approve now reminds you at the top of every settings page until the addresses are approved.

Expected result — only after saving in each section is the imported data validated and stored; Apply alone changes nothing on the server.

What an import does not carry

• API keys — no format contains them; add them per connection.
• Provider type — imported connections start as OpenAI-compatible; change it where needed.
• An agent's fixed model and the general values in an AnyLLM export (default model, AI access, rate limit) — set them again.
• The skill setting and agent assignment of templates — imported entries arrive as plain templates; tick Skill: run with one click in the chat (sends immediately) again and reassign the agent where needed.
• Over-long values are cut: connection name 40, endpoint 300, model name 100, agent instructions 8,000, template text 4,000 characters.
• Apply keeps at most 5 connections, 8 agents and 30 prompts (existing entries count) and drops empty connection cards.

Common problems

What you can recover — and what not

Related topics

• Administration & Configuration
• Opening the Settings and Initial Setup
• Setting Up LLM Connections
• Managing Endpoint Approvals
• Providing Agents, Prompts and Skills for Everyone
• General Settings: Language, Default Model, AI Access, Rate Limit, Site Knowledge
• Licence, Data Retention and Automatic Deletion
• Managing Chat History: Pin, Fork, Archive, Delete
• Troubleshooting
• FAQ