Most desktop LLM clients focus on chat; Proma reframes the desktop as an "Agent workbench" where short chats and multi-step, tool-driven Agent workflows coexist and persist. That shift makes it easier to take a conversational idea and turn it into a recoverable, permissioned task that runs locally or via bridged robots.
What Sets It Apart
- Local-first workspace model: conversations, agent sessions, skills and attachments are stored as JSON/JSONL under ~/.proma, enabling easy backups, inspection, and offline-friendly workflows. This matters when you want verifiable provenance or low-dependency installs.
- Agent-native architecture: built around the Claude Agent SDK and a Provider Adapter, Proma exposes Agent orchestration (sub‑agents/tasks, plan-confirmation, long-task streaming) rather than treating agents as single-turn chat wrappers — so multi-step automations can show their internal tool calls and state in the UI.
- Workspace-scoped Skills & MCP: each workspace can register Skills and an MCP server endpoint, making reusable capability bundles that are isolated per project. That reduces accidental cross-project data leakage and simplifies repeating complex tasks.
- Remote-robot bridges: native support to bridge a local Agent to Feishu (Lark) group chats and other IM platforms, so Agents running on your desktop can be triggered from mobile or group contexts.
Who It's For & Trade-offs
Great fit if you want a desktop-first Agent environment where data stays local, you need workspace isolation for repeated automations, or you need to trigger local Agents from chat platforms. Developers who use Bun/Electron/TypeScript will find the monorepo and tooling familiar. Look elsewhere if you need a fully hosted managed SaaS (Proma offers a separate commercial build) or if you require Agent channels that only speak OpenAI/Google protocols — Agent mode requires an Anthropic or Anthropic-compatible channel.
Where It Fits
Proma sits between lightweight chat clients and full cloud-hosted agent platforms: it is for users who prefer a desktop app that can escalate conversations into auditable, multi-step agent tasks, while keeping attachments and configs on-disk. Technically oriented teams that value inspectability, workspace-scoped skills, and bridged mobile triggers will get the most value.
