Selfship.ai

Selfship.ai is a service that turns failures in a production AI agent into reviewable pull requests. It reads customer conversations, works out what each customer was trying to do and whether they succeeded, and looks for patterns where many people hit the same problem. When it finds one, it pinpoints the moment the agent goes wrong and opens a pull request on GitHub with the proposed change, along with eval evidence such as impact, sample sessions and scores. The team reviews and merges the change. After a merge, Selfship watches real traffic to confirm the fix held. If it did not, it opens another fix pull request, up to a retry limit, and then marks the item as needing attention. It is aimed at engineering teams running customer facing agents. It works with any model and framework, is built on OpenTelemetry, and lists integrations with OpenAI, Anthropic, Gemini, LangChain, LlamaIndex, the Vercel AI SDK, CrewAI, Bedrock and Mistral. Analysis runs in ephemeral sandboxes, and the vendor states that code and traces are never used for model training. A public demo workspace with static sample data is available without signing up. Pricing is based on sessions and retention, not seats. The Base plan costs $20 per month after a 14 day trial with no card, and includes 6,000 sessions per month, 120 days of retention, and unlimited repositories, organizations and teammates. Extra packs of 6,000 sessions cost $20 each. Enterprise pricing is custom and adds SOC 2 Type II, HIPAA and GxP compliance, audit logs and dedicated support. A session is one agent turn.