{"id":"mng-dev-ai-agentrove","name":"agentrove","af_score":44.8,"security_score":52.5,"reliability_score":22.5,"what_it_does":"Agentrove is a self-hosted Claude Code workspace: a React web UI with a FastAPI backend that provisions per-workspace sandboxes (Docker or host), runs Claude Code via claude-agent-sdk/Claude Code CLI, and routes requests to Anthropic directly or through an Anthropic-compatible bridge for other providers (OpenAI/OpenRouter/GitHub Copilot/custom compatible endpoints). It also supports extension surfaces like MCP servers, skills/agents/commands, and streaming with SSE and cancellation.","best_when":"You want to run Claude Code capabilities on your own infrastructure (web IDE + sandboxed execution) and need provider switching/routing while keeping consistent workspace state and tool/MCP orchestration.","avoid_when":"You cannot secure JWT/API endpoints and sandbox/runtime secrets, or you require a stable API contract without anticipating breaking changes (the project states it is under active development).","last_evaluated":"2026-03-30T13:35:31.273049+00:00","has_mcp":false,"has_api":true,"auth_methods":["JWT (SECRET_KEY used for JWT signing per .env)","Provider tokens in UI/settings (Anthropic token; OpenAI device flow; GitHub Copilot device flow; OpenRouter API key; custom Anthropic-compatible base_url + token)","Basic use via browser session (implied by web UI; exact auth flow not specified in README)"],"has_free_tier":false,"known_gotchas":["Breaking changes are expected because the project is under active development.","Provider routing and sandbox execution introduce operational complexity; agent workflows may fail if sandbox runtime or bridge services are misconfigured.","Per-workspace sandbox provider is locked at creation time; switching may require recreating workspaces.","README mentions streaming SSE with cancellation and resumable events, but does not specify agent retry/resumption semantics or idempotency guarantees."],"error_quality":0.0}