Releases·Tool Release
LangGraph handles the agent loop. Model fallbacks, per-agent spend caps, keys and audit are a separate layer - here is how I split them out (LangChain points at it unchanged)
A developer released Agnos, an open-source (MIT) infrastructure layer that sits under LangGraph agents to handle model fallbacks, per-agent spend caps, key management, and audit trails. The post frames LangGraph as strong at the agent loop but lacking production concerns, and seeks community critique of the architecture.
7 engagement1 sourceMon, Sep 7, 2026, 02:20 PM
AI Overview
- The open-sourced Agnos layer (MIT) adds an OpenAI-compatible endpoint with ~1 ms median overhead, encrypted per-request key injection, and per-worker spend caps enforced below the graph
- Engines stay swappable behind a fixed port (built-in, LiteLLM's 100+ providers, Bifrost), with a test failing the build if core code references a specific engine
- Declarative guardrails block or redact secrets and PII across engines, and every call lands in one cost-attributed audit log
Related
Social·Community
LiteLLM/Bifrost gateways hold real provider keys in-process; a poisoned PyPI package drained keys from CI, pushing vault-injection containment
A Reddit post argues that LLM gateways such as LiteLLM and Bifrost, which hold real provider keys alongside admin panels and tool servers, pose a bigger security risk now that agents take actions rather than just generate replies. A compromise previously meant only a leaked key; now it can expose keys plus an agent with write access to databases and tools.
A Reddit post argues that LLM gateways such as LiteLLM and Bifrost, which hold real provider keys alongside admin panels and tool servers, pose a bigger security risk now that agents take actions rather than just generate replies. A compromise previously meant only a leaked key; now it can expose keys plus an agent with write access to databases and tools.
15 engagement1 sourceredditMon, Sep 7, 2026, 01:21 PM
Products·launched
Atom LLM Gateway: route and fallback between local and cloud LLMs
Atom is an open-source multi-agent platform with an LLM Gateway module that acts as a unified local proxy for routing and failover between local Ollama models and cloud APIs. It solves the problem of managing API switching, failover logic, and context limits for developers using local models alongside cloud endpoints in agent workflows, Cursor/Windsurf, or custom scripts.
Atom is an open-source multi-agent platform with an LLM Gateway module that acts as a unified local proxy for routing and failover between local Ollama models and cloud APIs. It solves the problem of managing API switching, failover logic, and context limits for developers using local models alongside cloud endpoints in agent workflows, Cursor/Windsurf, or custom scripts.
2 engagement1 sourceredditSun, Aug 2, 2026, 08:44 PM
Releases·Tool Release
AgentSwarms releases self-hostable agent platform with real LangChain/LangGraph notebooks
AgentSwarms-fyi released a self-hostable agent platform that runs real LangChain/LangGraph notebooks in sandboxed server kernels, with a bridge that returns a genuine BaseChatModel. The platform aims to simplify agent development while remaining compatible with existing LangChain tooling, and includes an honest comparison to using LangGraph directly.
AgentSwarms-fyi released a self-hostable agent platform that runs real LangChain/LangGraph notebooks in sandboxed server kernels, with a bridge that returns a genuine BaseChatModel. The platform aims to simplify agent development while remaining compatible with existing LangChain tooling, and includes an honest comparison to using LangGraph directly.
1 engagement1 sourceredditTue, Jul 28, 2026, 08:46 PM
Releases·Tool Release
Dev releases langgraph-openai-serve (LGOS) v0.16.0, serving self-hosted LangGraphs behind the OpenAI API
A developer released langgraph-openai-serve, an open-source tool that lets self-hosters expose LangGraph graphs behind an OpenAI-compatible API. It targets the deployment gap the author hit when running LangGraph on their own infrastructure, where managed options didn't fit an open-source, self-hosted stack.
A developer released langgraph-openai-serve, an open-source tool that lets self-hosters expose LangGraph graphs behind an OpenAI-compatible API. It targets the deployment gap the author hit when running LangGraph on their own infrastructure, where managed options didn't fit an open-source, self-hosted stack.
4 engagement1 sourceredditMon, Sep 7, 2026, 10:03 AM
Social·Community
Users seek open-source LLM gateway with dynamic runtime routing
A developer is asking for an open-source, self-hosted AI/LLM gateway that supports dynamic runtime routing and model management, to sit between agent frameworks and multiple LLM providers. They need to add/remove models, enable/disable them, and change routing weights without restarting, plus features like load balancing, fallbacks, retries, timeouts, health checks, and an OpenAI-compatible API. They've looked at LiteLLM and Portkey but find them lacking for simple self-hosting.
A developer is asking for an open-source, self-hosted AI/LLM gateway that supports dynamic runtime routing and model management, to sit between agent frameworks and multiple LLM providers. They need to add/remove models, enable/disable them, and change routing weights without restarting, plus features like load balancing, fallbacks, retries, timeouts, health checks, and an OpenAI-compatible API. They've looked at LiteLLM and Portkey but find them lacking for simple self-hosting.
31 engagement1 sourceredditThu, Aug 6, 2026, 05:01 PM