
A.I. Harness built in Rust

A.I. Harness built in Rust
Tendrite is a local-first command center for AI-assisted coding work. It provides a native workspace that integrates projects, agents, memory, source control, and decision tracking into a single environment. The tool is designed to help developers manage and guide AI agents through coding tasks without relying on chat-based interfaces. Tendrite operates primarily on the user's local machine, keeping code and data under the user's control while allowing connections to compatible AI model subscriptions, API keys, or local models. The workspace includes several core components. A memory substrate captures facts, lessons, warnings, and handoffs that agents learn during real work, distinguishing between in-flight information, durable knowledge, and history. It retrieves only context relevant to the current task, using keyword, semantic, and relationship signals. Near-duplicate observations reinforce existing knowledge rather than creating redundant entries. Users can review, consolidate, and retire stale knowledge, with the ability to pin important items, inspect the knowledge graph, and recover archived memories. Additional features include projects, memory, skills, artifacts, agents, missions, a campaign board, source control integration, an editor, and provider setup. The tool also offers a phone companion with a user's own network path, account-backed cloud entitlement, managed phone relay, and account device inventory with revocation. Typical use cases involve launching focused work sessions with defined roles for agents, such as research, implementation, or review. Users can turn a product goal into a bounded mission before agents begin changing files. A live map of the work shows the team's progress, blockers, and decision points. The workspace supports warm multi-project workspaces and a cross-project team brain, allowing context to persist across sessions. Per-agent model and account routing enable different agents to use different AI providers. Tendrite is currently in a limited beta while the developers harden operations.