LangChain has released Open SWE, an open-source framework for companies building coding agents that work inside existing engineering processes. It packages patterns seen in internal systems at Stripe, Ramp and Coinbase rather than requiring every team to assemble the same components from scratch.
Each task runs in a separate cloud sandbox with broad permissions inside that boundary. The agent can execute shell commands, fetch web pages, call APIs, commit code and open a draft GitHub pull request. It can also update Linear tickets and reply in Slack, while organizations can explicitly add their own tools.
Open SWE gathers repository rules from an `AGENTS.md` file and combines them with the full Slack thread or Linear issue that started the job. It can delegate subtasks to child agents with separate context, while deterministic middleware handles follow-up messages, tool errors and the final commit or pull request if the model omits that step.
The framework is a starting point, not a turnkey guarantee of safe code changes. Teams still need to choose sandbox providers, restrict credentials, review tools and test organization-specific workflows. Its practical value is making those controls configurable while keeping improvements to the underlying Deep Agents framework available without maintaining a private fork.