Sulcus, an agent runtime/control layer, launched a native LangGraph integration that surfaces execution data for complex multi-agent workflows and plans runtime safeguards. The founders are seeking feedback from developers running production LangGraph graphs.
Sulcus is a control layer for AI agents that consolidates visibility into agent runs, tool calls, LLM activity, errors, and inter-agent relationships in one place. It also provides runtime controls such as permissions, approvals, and limits. It targets developers running multiple agents built with frameworks like LangGraph and CrewAI who otherwise struggle to understand what their agents are doing.
Sulcus is a control layer for AI agents that consolidates visibility into agent runs, tool calls, LLM activity, errors, and inter-agent relationships in one place. It also provides runtime controls such as permissions, approvals, and limits. It targets developers running multiple agents built with frameworks like LangGraph and CrewAI who otherwise struggle to understand what their agents are doing.
OxDeAI is an open-source protocol that enforces a fail-closed authorization boundary before agent actions execute. It uses signed artifacts and adapters for LangGraph, CrewAI, and AutoGen to prevent retry amplification, budget leaks, stale-state executions, and permission drift. It solves the problem of agents performing unauthorized or non-idempotent actions for developers building production agent systems.
OxDeAI is an open-source protocol that enforces a fail-closed authorization boundary before agent actions execute. It uses signed artifacts and adapters for LangGraph, CrewAI, and AutoGen to prevent retry amplification, budget leaks, stale-state executions, and permission drift. It solves the problem of agents performing unauthorized or non-idempotent actions for developers building production agent systems.
Runkite, a self-hosted Agent Protocol control plane, is now available in preview under the Business Source License. It supports LangGraph, LangChain, CrewAI, LlamaIndex, AutoGen, and LangGraph.js with a unified runner protocol, plus an ops plane for durable threads, streaming, and an admin UI. Governance features like connector grants, audit, HITL, and kill are enforced on the plane for SQL backends.
Runkite, a self-hosted Agent Protocol control plane, is now available in preview under the Business Source License. It supports LangGraph, LangChain, CrewAI, LlamaIndex, AutoGen, and LangGraph.js with a unified runner protocol, plus an ops plane for durable threads, streaming, and an admin UI. Governance features like connector grants, audit, HITL, and kill are enforced on the plane for SQL backends.
TokenSentinel is an open-source library that instruments LLM clients to detect and block token-waste patterns in LangChain/LangGraph agents mid-session, including tool loops, retry storms, and context bloat. It provides inline waste signals and a callback for logging, alerting, or blocking, complementing rather than competing with LangSmith's tracing.
TokenSentinel is an open-source library that instruments LLM clients to detect and block token-waste patterns in LangChain/LangGraph agents mid-session, including tool loops, retry storms, and context bloat. It provides inline waste signals and a callback for logging, alerting, or blocking, complementing rather than competing with LangSmith's tracing.