Context
Also written: Context Window, Context Management, Execution context, Context Uniformly, Context-aware, Context engineering, Contextual Information, Shared context
The information an AI uses to understand and respond to your requests, including past conversation and provided data.
Context is the background information an AI model uses to understand what you're asking and give you a relevant answer. Think of it like a conversation: you remember what you just talked about, and the AI does too, up to a certain point.
This information can include your current prompt, previous messages in the chat, and any files or data you've uploaded. For example, if you ask an AI to summarize a document, the document itself is part of the context. The amount of context an AI can handle is often measured in tokens, which are like pieces of words.
Context in AI refers to the input data provided to a model to inform its generation process. This includes the current prompt, conversational history, and any external documents or data the model is given access to. The size of this context window, measured in tokens, is a critical parameter affecting the model's ability to maintain coherence and recall information over extended interactions or large datasets.
Larger context windows allow models to process more information at once, leading to better performance on tasks requiring long-range dependencies or extensive background knowledge. However, processing larger contexts can be computationally more expensive and may introduce challenges in effectively utilizing all the provided information.
Recent developments show a strong focus on enhancing AI's ability to remember and utilize past interactions and environmental details. Companies are building tools and APIs that allow AI agents to access and manage larger amounts of information, improving their performance on complex, multi-turn tasks and in specific operational environments.
- Evaluating AI agents: moving past the vibe checkGoogle · Google Developers Blog ·
- Build Plugins for ChatGPTOpenAI · OpenAI on YouTube ·
- Gemini API Cookbook: fix(get_started): traverse previous_interaction_id to export full multi-turn history (#1407)Google · Gemini API Cookbook ·
- How V7 gives AI agents institutional memoryOpenAI · OpenAI News ·
- Introducing the Agents APIOpenAI · OpenAI News ·
- Gemini API Cookbook: docs(quickstarts): enrich video understanding with agentic guidance, multi-turn, and custom FPS (#1358)Google · Gemini API Cookbook ·
- Projects in ChatGPTOpenAI · OpenAI Help Center ·
- Substrate-Aware AI Agents: Execution Context as a First-Class InputAnthropic · arXiv cs.AI ·
- How Perplexity Brought Voice Search to Millions Using the Realtime APIOpenAI · OpenAI Developers Blog ·
- 15 lessons learned building ChatGPT AppsOpenAI · OpenAI Developers Blog ·
- The Economics of Agent Optimization: Context engineering for enterprise AI agentsMicrosoft · Microsoft Azure Blog ·
- OpenAI Cursor split 💔, self-improving AI 🧠, GPT-6 Astra 🚀OpenAI · TLDR AI ·
- Quoting Drew BreunigAnthropic · Simon Willison ·
- Your New Backup Brain Is HereAnthropic · Matt Wolfe ·
- Grok API Quickstart: Get Your API KeyxAI · xAI Docs: Guides ·
- 3 things to know about the new CopilotKit integration for AngularGoogle · Google for Developers on YouTube ·
- What’s New in Microsoft 365 Copilot | July 2026Microsoft · Microsoft 365 Copilot Blog ·
- Text GenerationxAI · xAI Docs: Guides ·
- Using custom GPTsOpenAI · OpenAI News ·
- Introducing advanced tool use on the Claude Developer PlatformAnthropic · Anthropic Engineering ·
- Introducing OpenAI FrontierOpenAI · OpenAI News ·
- StreetReaderAI: Towards making street view accessible via context-aware multimodal AIGoogle · Google Research Blog ·
- Work smarter with your company knowledge in ChatGPTOpenAI · OpenAI News ·
- Can We Stop AI Deception? Apollo Research Tests OpenAI's Deliberative Alignment, w/ Marius HobbhahnOpenAI · The Cognitive Revolution ·
- Context Rot: How Increasing Input Tokens Impacts LLM Performance (Paper Analysis)Anthropic · Yannic Kilcher ·
- Accelerating engineering cycles 20% with OpenAIOpenAI · OpenAI News ·
- OpenAI CodexOpenAI · OpenAI News ·
- Anthropic Claude
- Anthropic research proposes substrate-aware AI agents that use execution context as a primary input.
- Google Gemini
- Google Gemini API updates allow for exporting full multi-turn chat histories and enriching video understanding with agentic guidance.
- OpenAI ChatGPT
- OpenAI is enabling AI agents to remember company history, build custom tools within ChatGPT, and offering an API for cloud agents with managed infrastructure.
- Prompt
- A prompt is the specific instruction or question given to the AI, while
- Training Data
- Training data is used to build the AI model itself, whereas conte
- Memory
- Memory in AI can refer to various mechanisms for storing information, wi
The concept of context in AI has roots in natural language processing and linguistics, referring to the surrounding text or situation that gives meaning to words or phrases. In modern large language models, it specifically refers to the input provided to the model for a given inference. Source ↗
Start a new chat with an AI assistant like ChatGPT or Claude. Ask it a question, then ask a follow-up question that relies on the previous answer. For example, ask 'What is the capital of France?' and then ask 'What is its population?'. Notice how the AI uses the context of your first question to answer the second.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- ToolSelf-Compacting Pi Agent CodebaseCode for an agent that controls its own context window and decides when to compact information. · 120 min
- Anti-patternAvoid cramming templates into the system promptStop putting templates and worked examples in the system prompt; use skills instead for better performance. · 10 min
- HeuristicSeparate context gathering from deep reasoning for AI agentsDelegate context curation to a separate agent to let the primary reasoning model focus on solving the task. · 15 min
- HeuristicManaging context window for long agent conversationsAgents must manage the context window to prevent prompt length from exceeding model limits. · 5 min
- HeuristicAI context window management for long tasksAI can 'forget' information when its context window fills; it needs to compact and note progress. · 1 min
Questions people ask
- How much context can an AI remember?
- The amount of context an AI can remember is limited by its context window, measured in tokens. This limit varies by model, with some newer models supporting much larger context windows than older ones.
- Can AI use external documents as context?
- Yes, many AI models can be given access to external documents or data. This allows them to use the information within those documents as context to answer questions or perform tasks.
First seen on Subvolts Aug 10, 2021. Updated Oct 15, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.