Skip to content

PRAXO-CORE vs Other Go Agent Frameworks

This page compares PRAXO-CORE with several widely used Go agent frameworks. It is a high-level orientation, not a benchmark. The comparison also covers fit with the AxisVela ecosystem, since PRAXO-CORE is the framework layer for that family. It is based on public documentation as of October 2026; upstream projects evolve quickly, so verify details before adopting.

PRAXO-CORE LangChainGo Eino Google ADK Go Microsoft Agent Framework Go
Primary focus Governed, embeddable agent runtime primitives LangChain-style composable LLM apps LLM app framework with graph orchestration and ADK Code-first agent toolkit, Gemini-optimized but model-agnostic Production agents and multi-agent workflows (public preview)
AxisVela ecosystem First-class seams for NOETIVELA, Mnemovela, Harmovela, GNOSIVELA, AEGIVELA, RHEOVELA, LIMENORA Custom adapters/tools; no AxisVela-specific seams Custom adapters/tools; no AxisVela-specific seams Custom adapters/tools; no AxisVela-specific seams Custom adapters/tools; no AxisVela-specific seams
Orchestration Single-agent loop, middleware, subagents, flow Chain/Parallel Chains, agents, tools Graphs/workflows, ReAct ChatModelAgent, DeepAgent multi-agent Workflow agents, graph workflows, multi-agent composition Graph workflows: sequential, concurrent, group, routing, subworkflows
Providers Provider-neutral model.Model; NOETIVELA/PRAXOVELA adapters Broad built-in provider set Extensions via eino-ext Gemini first; OpenAI adapter; model-agnostic Foundry, Azure, OpenAI, Anthropic, Gemini, Copilot
Tools and MCP Tool contracts, JSON Schema, retry, gateway; Janus/LIMENORA gateway; MCP client is consumer-owned Tools; MCP not first-class Tools; MCP via eino-ext Rich tools, MCP toolset, A2A Tools, MCP, A2A, AG-UI
Memory and RAG Interfaces and retrieval contracts; Mnemovela memory, GNOSIVELA retrieval Memory, loaders, vector stores Retriever and context components Sessions, memory, artifacts Memory, text search/RAG, vector stores
Governance and extension Policy hooks, AEGIVELA policy, path containment, sandbox seam, checkpoints, HITL, process/WASM plugin plane, event sinks Callbacks HITL interrupt/resume, callbacks Lifecycle plugin hooks, auth, telemetry Checkpointing, tool approval/HITL, OpenTelemetry
License MIT MIT Apache-2.0 Apache-2.0 MIT

AxisVela Ecosystem Fit

PRAXO-CORE is the only framework in this table designed as the AxisVela framework layer. It exposes seams for:

  • NOETIVELA for governed model routing and inference
  • Mnemovela for memory
  • Harmovela for events and coordination
  • GNOSIVELA for semantic retrieval and grounding
  • AEGIVELA for authorization
  • RHEOVELA for graph and workflow execution
  • LIMENORA for external tools, APIs, and MCP gateways

These are contracts, not bundled adapters; products supply the concrete clients. Other frameworks can integrate with AxisVela through custom adapters, tools, HTTP, or MCP, but those integrations are not part of their core contracts. See AxisVela ecosystem integration.

Positioning

  • PRAXO-CORE is deliberately lower-level. It does not bundle provider SDKs, vector databases, or an MCP client. Its value is a small, MIT-licensed governance and runtime boundary that products can embed, with first-class seams for the AxisVela ecosystem.
  • LangChainGo is the broadest LangChain-style port in Go, with a large set of providers, chains, memory backends, document loaders, and vector stores.
  • Eino combines graph/workflow composition with an ADK, ReAct agents, and multi-agent patterns. Provider and tool integrations are largely delivered through eino-ext.
  • Google ADK Go is a code-first toolkit optimized for Gemini but usable with other models. It includes workflow agents, graph workflows, sessions, memory, artifacts, MCP tools, and A2A.
  • Microsoft Agent Framework Go targets production agents and graph-based multi-agent workflows, with checkpointing, HITL, OpenTelemetry, MCP, and A2A. The Go implementation is in public preview.

When to Choose PRAXO-CORE

Choose PRAXO-CORE when you need:

  • native seams for the AxisVela ecosystem: NOETIVELA, Mnemovela, Harmovela, GNOSIVELA, AEGIVELA, RHEOVELA, or LIMENORA
  • a stable, small runtime boundary to embed in a product
  • provider-neutral contracts with product-owned adapters
  • policy, path containment, sandbox seams, checkpoints, and HITL in the core
  • a process/WASM plugin plane with a conformance kit
  • an MIT-licensed framework with no AGPL imports

When to Choose Another Framework

  • Choose LangChainGo for a familiar LangChain-style API and broad built-in integrations, especially for RAG-heavy applications.
  • Choose Eino for graph/workflow orchestration and a mature multi-agent ADK.
  • Choose Google ADK Go for a Gemini-first, cloud-native agent toolkit with strong deployment and multi-agent support.
  • Choose Microsoft Agent Framework Go when you want graph workflows and checkpointing inside the Microsoft/Foundry ecosystem.

If you need deep AxisVela integration, start with PRAXO-CORE rather than adding product-specific adapters to a general-purpose framework.

Notes

This comparison does not score performance, stability, or ecosystem maturity. Feature descriptions follow each project's public documentation and may lag behind the latest releases. Open an issue if a comparison point is outdated.