BotwarrenBotwarren / Integrations
Integrations
The ship agent watches your AI coding tools through small adapters. Each adapter declares exactly what it reads, and a redaction test plants unique "canary" text in every field an adapter could touch and fails the build if that text ever reaches an event, a public projection or a log. Here is what works today, and how well each is tested.
| Tool | What the adapter reads | How well tested |
|---|---|---|
| Claude Code | Metadata of its local session logs: file names and times, the project folder, tool kind, a failure flag, model family, token counts reduced to a gauge. | Verified with real logs Agent test suite runs on Linux, macOS and Windows. |
| Codex | Rollout file names and times, and only the id, folder, originator and source of the first line of each. | Verified with real logs |
| Herdr | The list of agents and a ping over Herdr's local socket; it never ends a session while Herdr is unreachable. | Fixture-tested Recorded protocol shapes; Windows not verified. |
| Ninox | Session status and activity from Ninox's local API on 127.0.0.1; names, repos, summaries and costs are always dropped. macOS and Linux only. | Fixture-tested Not yet checked against a live Ninox install. |
Experimental
Adapters for OpenCode and Gemini CLI exist and are tested only against their documentation (they are not verified live). Their sessions appear on a ship as generic crew members. Run botwarren adapters to see which adapters your agent has switched on and botwarren adapters explain <id> for the exact list of what one reads.
Sessions from different tools that belong together are merged into one crew member, so a Claude Code session started through Herdr or Ninox is not drawn twice.
Botwarren is an independent product of TurboKast Ltd. It is not made, endorsed or sponsored by Anthropic or OpenAI; Claude Code and Codex are their products.