Effective 2026-10-15. Applies to the Agents Deck plugin for JetBrains IDEs (dev.agentsdeck.jetbrains), published by Agents Deck.
Agents Deck has no analytics service or custom telemetry endpoint. Local session browsing and diff review do not send your code or prompts to us. Optional diagnostic reports use the IDE's JetBrains reporting service and can be shared with us as the plugin developer (§3.6).
The data flows that leave your machine are listed in §3, with agent CLI traffic explained in §4.
Agents Deck is a reader of files the Claude Code and Codex command-line tools already write. It reads, on your computer only:
~/.claude/projects/<encoded-cwd>/<session-uuid>.jsonl and the Codex equivalent. These contain your prompts, the agent's replies, tool calls, and excerpts of files the agent read or wrote.~/.claude/file-history/<session-id>/…, used to diff and revert what an agent changed.~/.claude/history.jsonl.~/.claude.json and related settings, including the env blocks of configured MCP servers (see §4).Reading these files does not transmit them to us. Information you include in feedback or an IDE diagnostic report may be shared as described in §3.5–3.6.
claudeAgents.xml and siblings). Choices and layout, not content.~/.claude/claude-agents.json.These flows depend on the features you use and, for IDE diagnostics, your reporting consent and settings.
To show token usage, plan limits and the Data privacy control in Settings › About, the plugin calls api.anthropic.com using the OAuth token Claude Code already stored on your machine for your own account. It sends no prompt text and no code — only the authenticated request itself.
The Data privacy control writes exactly one field on your account (grove_enabled) when you click it. That control writes only that field.
Your saved Cursor API key stays in the IDE's PasswordSafe. The plugin sends it only to https://api.cursor.com as an Authorization header; redirects are refused. While Sessions is visible, it reads the first 50 cloud agents on open and once per minute. While a Cursor cloud chat is visible, it reads that chat's runs and streams replies and tool events; hiding it closes the stream. Models are fetched when requested and cached for one hour.
Starting a cloud agent sends your prompt, selected model and credential-free repository origin URL to Cursor. Follow-up, Stop and Archive send requests for that agent. Cursor executes the work remotely under your account; its own policies apply. Cloud replies and list/model caches stay in memory, are cleared when the key changes or is removed, and are not written to the plugin's transcript store. Browser-only CLI sign-in does not enable these API calls.
If you have YouTrack, TeamCity or GitHub credentials exported in your environment or present in an MCP server's env block, the plugin uses them to show issue and build status. Requests go to the host you configured — nowhere else. The plugin never stores these tokens; it reads them where your other tooling already keeps them.
If you install an Agents Deck extension by naming a GitHub repository, the plugin contacts api.github.com and downloads that release's plugin archive. This happens only when you ask for a specific repository.
An extension is a JetBrains plugin, and once installed it runs with the same access as the IDE itself — your files, your credentials and your network. Nothing downloaded this way has been reviewed by JetBrains or by us. Before anything is installed, the plugin checks the archive's signature and shows you what it found, together with where the archive came from; an archive whose signature does not match its contents is refused outright. Installing is always your decision, and declining deletes the downloaded copy.
Off by default. No socket is opened until you enable it and pair a device.
When enabled, your IDE serves your paired phone over HTTPS. The phone pins the server's public key at pairing and refuses any mismatch; every request carries a bearer token whose digest alone is stored; repeated pairing failures lock a peer out. The socket binds to loopback only — it is reachable from your local network solely for interfaces you explicitly tick.
Optionally, you can make the bridge reachable from outside your network through a public TCP relay (bore, a separate tool you install yourself — the plugin bundles no binaries). The relay forwards encrypted bytes: your phone's TLS session runs end to end to your IDE, so the relay operator cannot read what passes through it. Data reaching your phone this way includes session transcripts, which contain your source code — enable this only if that is acceptable to you.
The About page's feedback and problem-report actions open your own mail client with a prepared message. Nothing is sent until you review it and press send in your mail client.
Agents Deck uses JetBrains Marketplace Exception Analyzer for IDE diagnostic reports. JetBrains receives submitted reports and makes them available to us to investigate and fix plugin problems. A report may include an exception message and stack trace, IDE/plugin/OS and Java runtime versions, update settings, recent IDE actions, a pseudonymous installation identifier, your comment, and included attachments. On IDE versions that support plugin freeze reporting, attachments may include thread dumps and freeze diagnostics. Messages and attachments can contain file paths or other sensitive context.
Reporting follows the IDE's consent flow and reporting settings. Review the report and selected attachments before submitting; newer IDE versions may also offer automatic reporting settings. Agents Deck adds no separate background reporting service. JetBrains' handling is governed by its Privacy Policy.
So model names and cost estimates cover models released after your build, the plugin downloads LiteLLM's public price list from raw.githubusercontent.com (GitHub) and our own model list (model-facts.json) from this site on github.io (GitHub Pages), each at IDE start and at most once a day, and keeps the files under the IDE's system directory. The requests are anonymous GETs with no account, prompt, code or identifier — GitHub sees your IP address, as it would for any download. Turn them off in Settings › Spending › Refresh model prices and details daily; the plugin then uses only what it shipped with and no longer applies the downloaded files.
Agents Deck runs the claude and codex binaries you installed, under your account. When you send a chat message, that CLI transmits your prompt and whatever project context it decides to include to Anthropic or OpenAI, under your existing agreement with them and their privacy policy — exactly as it would if you ran it in a terminal.
The plugin does not add to, alter, or copy that transmission, and does not send anything to those providers on its own behalf beyond §3.1. If you want to know what your agent sends, that is a question for your agent vendor's policy, not this one.
Agents Deck extensions (Linear, Sentry, Granola, Tessl) are separate plugins you install separately. When installed and configured with your own API key, each one contacts its own vendor's API on your behalf. Keys you type into an extension's setup dialog are stored in the IDE's password safe (your OS keychain), never in plain settings files and never in logs. Each extension's data flow is its own; this policy covers the Agents Deck plugin.
The plugin is a developer tool and is not directed at children.
The plugin's local data remains under your control. For access, correction, export or deletion requests concerning feedback or diagnostic reports shared with us, contact allenkazoloto@gmail.com, including the report reference if available. For data held by JetBrains or another service provider, contact that provider under its privacy policy.
If a security issue in the plugin ever causes data to be exposed, we will publish it in CHANGELOG.md and notify affected users through the Marketplace listing, and will meet any notification duty applicable law places on us.
Material changes are dated here and summarised in CHANGELOG.md. The effective date at the top of this file is the version in force.
allenkazoloto@gmail.com — https://alenkazoloto.github.io/agents-deck