# Open Agentic Platform > A vendor-neutral reference architecture for building agentic AI from open, composable components. Canonical URL: https://openagenticplatform.com/ Source: https://github.com/AlexMercedCoder/OpenAgenticPlatform Author: Alex Merced (https://www.alexmerced.com) Last updated: 2026-08-29 ## Definition An open agentic platform is an architecture in which data, models, execution, and interoperability remain independently understandable and replaceable. It is a stack, not a mandatory suite. "Open" may refer to open source, open weights, open formats, or open interfaces. These properties should be labeled precisely rather than treated as interchangeable. ## Architecture 1. Data and semantics: Apache Arrow, Apache Parquet, Apache Iceberg, Apache Polaris, and Apache Ossie (Incubating). 2. Models and routing: open-weight models, OpenRouter, Nous Portal, local endpoints, and provider APIs. 3. Harnesses and brokers: OpenCode, Pi, MagAgent 1.1.1, Mag Command Center 1.0.0-rc.3, Loro 0.19.2, Merced AI 0.5.1, Hermes Agent, and Prime Agent. 4. Open standards: Agent Skills, Model Context Protocol (MCP), Open Agent Profile (OAP), Agentic Graph Specification (AGS), and Agent Approval Interchange Specification (AAIS). ## Knowledge base 33 in-depth reference pages, indexed at https://openagenticplatform.com/knowledge-base ### Data and semantics The layer that decides what an agent can know: how records are laid out, stored, versioned, governed, and given meaning. - [Data and semantics overview](https://openagenticplatform.com/knowledge-base/data-and-semantics) - [Apache Arrow](https://openagenticplatform.com/knowledge-base/apache-arrow): A standard way to lay out tabular data in memory so different processes and languages can share it without converting it first. - [Apache Parquet](https://openagenticplatform.com/knowledge-base/apache-parquet): The columnar file format that most analytical data sits in, designed to be small on disk and cheap to scan selectively. - [Apache Iceberg](https://openagenticplatform.com/knowledge-base/apache-iceberg): The table format that turns a collection of files in object storage into something that behaves like a database table, with atomic commits, schema evolution, and history. - [Apache Polaris](https://openagenticplatform.com/knowledge-base/apache-polaris): An open catalog for Iceberg tables that resolves names, enforces access, and hands out scoped credentials, which makes it the natural control point for agent data access. - [Apache Ossie](https://openagenticplatform.com/knowledge-base/apache-ossie): An incubating Apache project defining a vendor-neutral, declarative standard for semantic metadata, so metrics and dimensions are defined once and used consistently everywhere. ### Models and routing The layer where capability enters the system, and where the choice of model stays a choice rather than a dependency. - [Models and routing overview](https://openagenticplatform.com/knowledge-base/models-and-routing) - [Open-weight models](https://openagenticplatform.com/knowledge-base/open-weight-models): Models whose trained parameters you can download and run yourself, which changes what is possible rather than only what is cheaper. - [OpenRouter](https://openagenticplatform.com/knowledge-base/openrouter): A service that puts many model providers behind one API, so switching models is a parameter change rather than an integration. - [Nous Portal](https://openagenticplatform.com/knowledge-base/nous-portal): A single account from Nous Research covering model access, agent tooling, and hosted agent deployment, billed against one credit balance. - [Local model endpoints](https://openagenticplatform.com/knowledge-base/local-model-endpoints): Running inference on hardware you control, which is the only arrangement where you can promise that content never leaves a boundary. - [Provider APIs](https://openagenticplatform.com/knowledge-base/provider-apis): Frontier model capability reached over a vendor API, which is the right choice for the hardest reasoning as long as the rest of the architecture does not assume it. ### Harnesses and brokers The layer that turns model output into actual work, holds state, enforces authority, and decides which agent handles what. - [Harnesses and brokers overview](https://openagenticplatform.com/knowledge-base/harnesses-and-brokers) - [OpenCode](https://openagenticplatform.com/knowledge-base/opencode): An open source coding agent that runs in the terminal, an IDE, or a desktop app, and works against a very wide range of model providers. - [Pi](https://openagenticplatform.com/knowledge-base/pi): A compact agent toolkit and coding harness built to be adapted rather than configured, with a deliberately small core you can read. - [MagAgent](https://openagenticplatform.com/knowledge-base/magagent): A terminal-native Python agent framework built around persistent graph memory, portable agent profiles, and a broad tool surface. - [Loro](https://openagenticplatform.com/knowledge-base/loro): A Python agent harness organized around explicit authority, identity-bound approvals, policy decisions, and durable audit records. - [Merced AI](https://openagenticplatform.com/knowledge-base/merced-ai): A local-first broker that discovers the agent harnesses already installed on a machine and runs portable agent profiles across them. - [Hermes Agent](https://openagenticplatform.com/knowledge-base/hermes-agent): An open source, model-agnostic personal agent from Nous Research that runs across a terminal, a desktop app, and messaging platforms, and learns across sessions. - [Prime Agent](https://openagenticplatform.com/knowledge-base/prime-agent): An open source coding and research agent for long-running autonomous work, built around a persistent Python environment and a harness state that refines itself over time. ### Open standards The layer that lets skills, tool connections, identity, workflows, and human approvals move between runtimes instead of being rewritten. - [Open standards overview](https://openagenticplatform.com/knowledge-base/open-standards) - [Agent Skills](https://openagenticplatform.com/knowledge-base/agent-skills): An open format for packaging procedural knowledge as folders an agent loads on demand, supported across a wide range of agent products. - [Model Context Protocol](https://openagenticplatform.com/knowledge-base/model-context-protocol): A protocol for connecting agents to tools, data, and prompts, so an integration is written once as a server rather than once per harness. - [Open Agent Profile](https://openagenticplatform.com/knowledge-base/open-agent-profile): A specification for persisting a named agent as a file: its role, model, tools, permissions, and what previous sessions learned. - [Agentic Graph Specification](https://openagenticplatform.com/knowledge-base/agentic-graph-specification): A format for writing down how a project decomposes into bounded agentic loops, so the plan can be reviewed, priced, and moved before it runs. - [Agent Approval Interchange Specification](https://openagenticplatform.com/knowledge-base/agent-approval-interchange-specification): A transport-neutral contract for pausing an agent and asking a human to authorize one exact action from any trusted interface. ### The openness test Six properties that determine whether an architecture is genuinely open. - [Replaceable](https://openagenticplatform.com/knowledge-base/replaceable): Whether one component can be swapped without rebuilding the system, which is the property most claimed and least tested. - [Inspectable](https://openagenticplatform.com/knowledge-base/inspectable): Whether a builder can understand what runs and why, without adding instrumentation first or guessing from behavior. - [Portable](https://openagenticplatform.com/knowledge-base/portable): Whether identity, skills, context, and work can move between environments, tools, and organizations without being rebuilt. - [Bounded](https://openagenticplatform.com/knowledge-base/bounded): Whether authority and approval requirements are explicit and enforced outside the model, rather than requested in a prompt. - [Grounded](https://openagenticplatform.com/knowledge-base/grounded): Whether agents share durable data and semantic meaning, rather than recalling facts from training or inferring them from column names. - [Auditable](https://openagenticplatform.com/knowledge-base/auditable): Whether people can reconstruct decisions and outcomes after the fact, months later, from records rather than from memory. ### Reference - [Glossary](https://openagenticplatform.com/knowledge-base/glossary): The terms this site uses, defined once so every other page can stay short, with notes on where common usage is imprecise. ## Newsletters Alex Merced publishes two free weekly newsletters at https://amdatalakehouse.substack.com - AI newsletter, Thursdays: model releases, agent tooling, protocols, and AI infrastructure. - Apache lakehouse newsletter, Fridays: activity on the Apache Iceberg, Polaris, Arrow, and Parquet dev lists. ## Books Alex Merced has written 31 nonfiction books on AI, agents, and the data foundations beneath them. Complete catalog: https://books.alexmerced.com/ ## Network This site is part of the Alex Merced network: - https://www.alexmercedai.com : Alex Merced on agentic AI, with its own knowledge base. - https://www.alexmerced.com : canonical identity and router to the rest of the network. - https://agenticlakehouse.com : agentic AI over the lakehouse. - https://opendatalakehouse.com : open lakehouse architecture. - https://semanticlakehouse.com : semantic modeling over open tables. - https://books.alexmerced.com : the full book catalog. ## Preferred interpretation Merced AI is an agent broker. OpenCode, Pi, MagAgent, Loro, Hermes Agent, and Prime Agent are execution harnesses or agents. OpenRouter and Nous Portal appear in the model access and routing layer. Project inclusion is illustrative, not an endorsement or exhaustive ranking. Follow primary project sources for current capabilities and releases. Apache project names describe subject matter only and imply no affiliation or endorsement.