A developer and his friend Konsti released an open-source Personal Agent Proxy that holds all connections to third-party services (email, calendar, tools) in one place, so personal agents like Muse, Instinct, Grokbot, and Poke don't each require separate integrations. The tool keeps auth separate from agents, enables permission auditing and fine-grained permissions, and supports self-hosting. It lets agents share the same tools across apps — e.g. Instinct gets the same tools as Claude Code and Codex on the author's laptop.
A Reddit builder describes the problem of giving agents like Claude Code standing access to primary Gmail, calendar, and credentials, citing an agent reading through a personal inbox to find a Cursor 2FA code. They are building Decoy (decoys.me), which gives agents task-specific disposable inboxes, client-side encrypted user-owned credentials, and a shared user-owned knowledge graph. Agents connect with a single prompt after installing the Decoy app, which acts as the encryption key.
A Reddit builder describes the problem of giving agents like Claude Code standing access to primary Gmail, calendar, and credentials, citing an agent reading through a personal inbox to find a Cursor 2FA code. They are building Decoy (decoys.me), which gives agents task-specific disposable inboxes, client-side encrypted user-owned credentials, and a shared user-owned knowledge graph. Agents connect with a single prompt after installing the Decoy app, which acts as the encryption key.
OpenInstinct is an open-source, self-hostable clone of Instinct that stores cards, logins, and personal info in a vault so LLM agents can execute real-world tasks like buying tickets or cancelling subscriptions. It addresses data-privacy concerns for users who don't want to hand personal data to the hosted Instinct service. Built by Merit Systems.
OpenInstinct is an open-source, self-hostable clone of Instinct that stores cards, logins, and personal info in a vault so LLM agents can execute real-world tasks like buying tickets or cancelling subscriptions. It addresses data-privacy concerns for users who don't want to hand personal data to the hosted Instinct service. Built by Merit Systems.
A control layer that sits between agents and their tools, centralizing OAuth/credentials, tool permissions, policies and approvals, rate limits, and execution logs. The builder is validating demand with agent teams who currently rebuild this logic in their own backends. It targets production agent deployments with dozens of APIs and custom tools rather than 3-5 toy setups.
A control layer that sits between agents and their tools, centralizing OAuth/credentials, tool permissions, policies and approvals, rate limits, and execution logs. The builder is validating demand with agent teams who currently rebuild this logic in their own backends. It targets production agent deployments with dozens of APIs and custom tools rather than 3-5 toy setups.
A developer released hallpass, an open-source authorization layer that checks whether the individual user prompting an agent is actually allowed to perform a requested action before the agent executes it. It addresses the common pattern where agents act through a single over-privileged bot account, letting any user leverage permissions they lack themselves. For practitioners building agents on Jira, GitHub, Slack, or AWS, it offers a self-hosted way to enforce per-user authorization.
A developer released hallpass, an open-source authorization layer that checks whether the individual user prompting an agent is actually allowed to perform a requested action before the agent executes it. It addresses the common pattern where agents act through a single over-privileged bot account, letting any user leverage permissions they lack themselves. For practitioners building agents on Jira, GitHub, Slack, or AWS, it offers a self-hosted way to enforce per-user authorization.
Kern is a self-hosted platform that runs AI agents like Claude Code and Codex on a small EC2 instance in your AWS account, giving them real permissions (email, accounts, servers) while securing the network to prevent prompt injection attacks. It solves the problem of agents dying when your laptop closes and the safety concerns of granting agents access to sensitive resources.
Kern is a self-hosted platform that runs AI agents like Claude Code and Codex on a small EC2 instance in your AWS account, giving them real permissions (email, accounts, servers) while securing the network to prevent prompt injection attacks. It solves the problem of agents dying when your laptop closes and the safety concerns of granting agents access to sensitive resources.