# OpenSymbolicAI > OpenSymbolicAI is an open-source framework that brings software engineering practices like typed primitives, explicit decompositions, and a PlanExecute architecture to AI agent development. It treats AI behavior as code, not prompt strings. OpenSymbolicAI separates planning from execution. The LLM writes a plan; Python runs it. Agents built with it are reliable, debuggable, and maintainable. The core runtime is open source. ## Core Concepts - [Homepage](https://www.opensymbolic.ai): Overview of the framework and its approach to AI engineering - [Tutorials](https://www.opensymbolic.ai/tutorials): Hands-on guides from first agent to production patterns - [About](https://www.opensymbolic.ai/about): Team and background - [About](https://www.opensymbolic.ai/about): Team and contact ## Blog - [DesignExecute: When Straight-Line Plans Aren't Enough](https://www.opensymbolic.ai/blog/design-execute): PlanExecute forbids loops and conditionals on purpose. DesignExecute adds them back, with guardrails, for the problems that actually need control flow. Here's when to reach for it, and what stays the same. - [Third Language, Same Result: MultiHopRAG in Go](https://www.opensymbolic.ai/blog/multihop-rag-go): Go joins Python and C# on the MultiHopRAG benchmark. Different runtime, different vector store, single static binary. Accuracy: 81.6%. The framework holds. - [Change Everything, Change Nothing: MultiHopRAG in Python and C#](https://www.opensymbolic.ai/blog/multihop-rag-cross-language): We swapped the language, the vector store, the code executor, and the type system. Accuracy moved by 0.9pp. The framework is the invariant, not the infrastructure. - [Agent-to-Agent Is Just Function Calls](https://www.opensymbolic.ai/blog/agent-to-agent): Multi-agent systems don't need new infrastructure. They use the same patterns that connect microservices today: typed interfaces, explicit wiring, and the auth and observability stack you already have. - [TravelPlanner Benchmark: 97.9% on 1,000 Tasks Where GPT-4 Gets 0.6%](https://www.opensymbolic.ai/blog/travelplanner-benchmark): OpenSymbolicAI achieves near-perfect scores on all 1,225 TravelPlanner tasks, outperforms LangChain and CrewAI head-to-head, and maps the model landscape across 11 LLMs and 4 providers. - [The Anatomy of PlanExecute: Why It Is What It Is](https://www.opensymbolic.ai/blog/plan-execute-anatomy): A deep dive into the design decisions behind OpenSymbolicAI's core blueprint: why plans are code, why execution is sandboxed, and why the planner only gets called once. - [English, Spec, or Code: How You Talk to the LLM Decides How Far You Get](https://www.opensymbolic.ai/blog/the-prompt-spectrum): Plain English gets you to 50% on day one. Specs get you to 70%. Code gets you to 99.999%. The question is which ceiling you're willing to live under. - [Secure by Design: How OpenSymbolicAI Makes AI Agents Safe](https://www.opensymbolic.ai/blog/security-by-design): Why security in OpenSymbolicAI is architecturally guaranteed, not hoped for. Introducing the Symbolic Firewall. - [Illustration Part 3: Cost & Reliability](https://www.opensymbolic.ai/blog/illustration-cost-reliability): Tool-calling costs 2.3x more and fails 20% of the time. Here's the math. - [Illustration Part 2: Token Economics](https://www.opensymbolic.ai/blog/illustration-token-economics): Tool-calling re-reads everything on every call. Here's exactly where the tokens go. - [Illustration Part 1: The Attention Loss Problem](https://www.opensymbolic.ai/blog/illustration-attention-loss): We gave a tool-calling agent detailed instructions. It ignored them. Here's why. - [LLM Attention Is Precious: Why ReAct Wastes It](https://www.opensymbolic.ai/blog/llm-attention-is-precious): A visual comparison of token usage between ReAct (Reasoning+Acting) and PlanExecute approaches. - [Behaviour Programming vs. Tool Calling: Two Paradigms for AI Agents](https://www.opensymbolic.ai/blog/behaviour-programming-vs-tool-calling): Why teaching agents through executable examples beats massive prompts with chain-of-thought reasoning, illustrated through a RAG agent implementation. - [Closing the Flywheel in Practice](https://www.opensymbolic.ai/blog/closing-the-flywheel-in-practice): A hands-on walkthrough showing how primitives and decompositions create the compounding improvement loop that prompt engineering lacks. - [The Missing Flywheel in Agent Building](https://www.opensymbolic.ai/blog/the-missing-flywheel-in-agent-building): Why language models keep getting better while AI agents remain stubbornly brittle, and how to fix it with structural separation of concerns. - [The OpenSymbolicAI Manifesto](https://www.opensymbolic.ai/blog/manifesto): How do we build non-deterministic intelligence on top of a deterministic foundation, without compromising either? ## Optional - [Illustration Part 1: The Attention Loss Problem](https://www.opensymbolic.ai/blog/illustration-attention-loss): Visual explanation of why tool-calling agents ignore detailed instructions - [Illustration Part 2: Token Economics](https://www.opensymbolic.ai/blog/illustration-token-economics): Where tokens go in tool-calling vs behaviour programming - [Illustration Part 3: Cost & Reliability](https://www.opensymbolic.ai/blog/illustration-cost-reliability): Cost and failure rate comparison between approaches