loomcycle
Loomcycle is a Go-based agentic runtime that runs alongside applications via HTTP/gRPC/MCP interfaces. It provides hardened agent loops, multi-provider routing (Anthropic, OpenAI, DeepSeek, Gemini, Ollama), and native MCP support with tenant isolation.
Ficha actualizada 23 sept 2026, 5:31 · upstream 23 sept 2026, 5:02 · release v1.90.0
Qué es loomcycle
Descripción editorial construida a partir del repositorio, su documentación y sus señales de mantenimiento.
Loomcycle serves as the substrate for agentic systems — a single Go binary (~50 MB) that orchestrates an agent loop across multiple LLM providers. It is designed to be configurable as either a managed sandbox or a full-featured dev environment, offering MCP-native communication and multi-replica HA. The runtime handles tool execution, memory management (vector, SQL, agentic), document storage with semantic search, context compaction, and multi-agent coordination through state hand-offs.
- Run agent workloads alongside existing applications without modifying their codebase
- Provide a hardened sandbox for untrusted prompts in shared server environments
- Develop agentic workflows locally using the same binary across multiple providers
- Deploy multi-tenant agentic systems with per-principal isolation and cost attribution
- Build agent orchestration tools that integrate natively with existing tooling (n8n, OpenAI Gateway)
- Run local development environments with full MCP support for rapid prototyping
- Single Go binary (~50 MB) running the entire agent loop including routing, memory, and tools
- Native MCP server (loomcycle mcp) that can be called from any tool without vendor SDKs
- Multi-replica HA with Redis-based session management and DB-backed locks
- Tenant isolation at both state and definition planes with per-tenant budgets and data retention
- Semantic search over chunked documents with inline Mermaid diagrams and image captions
- Context compaction that preserves reasoning traces while reducing memory footprint
- No built-in tool execution sandbox — agents can only call tools via HTTP/gRPC; no isolation from the host process
- Single binary architecture means all components (provider, agent, memory) are bundled together with a single build target
- Multi-tenant cost attribution requires per-call token tracking which adds overhead to each request
- Context compaction is manual or auto-triggered — there is no guaranteed automatic retention of long-running agents without intervention
- No built-in agent reasoning trace visualization for human review, only structured state snapshots
Perfil de infraestructura
Estimación inicial derivada del repositorio y de los servicios detectados.
Despliegue detectado
Qué señales reales de instalación hemos encontrado.
Install via Homebrew (macOS/Linux) or Docker (v0.11.2+). Use `loomcycle init --with-token` to create a config directory and auth token. For production, use the multi-arch Docker image with PostgreSQL/Redis persistence. Enable Bash for local development but restrict it to localhost only in sandbox mode. Always validate environment variables before starting; the runtime will fail on missing keys or invalid tokens.
Stack y dependencias
Componentes inferidos a partir de manifests, configuración y documentación.
Puede sustituir a
Relaciones detectadas por el análisis del repositorio y su categoría.
Experiencia real de la comunidad
Ayuda a contrastar la estimación automática con instalaciones reales.
Todavía no hay datos comunitarios para este proyecto.