# AGNT

> AGNT, the local-first agent OS: a desktop platform for building and running AI
> agents and automations. Agents run on the user's own machine, credentials stay
> in a local vault, and every run produces a receipt showing what the agent did
> and why - including the runs where it decided to do nothing.
> Not to be confused with AGNT ECO (Wikidata Q141609370); this AGNT is Q141905400.

AGNT is free to download and run (Community Core). Workflows are built from
nodes on a canvas; agents are given a job in plain language and the tools to do
it. There are 190 node types covering triggers, model providers, integrations
and control flow.

This file lists the pages on agnt.gg that answer a specific question, so a
retrieval system can go straight to the relevant one. Every URL below is a real
page on this site, generated from the same data as this index.

Research, documentation and whitepapers come first: those are the pages worth
quoting. The integration, provider and glossary sections below them answer
narrower "does AGNT do X" questions.

Not covered here: the account, login and billing pages, which need a session
and say nothing useful; and the marketplace listings, which are user-submitted
and change without notice. A full RSS feed of the research is at
https://agnt.gg/feed.xml.

Canonical host: https://agnt.gg

## Research and field notes

- [AI Harness Engineering: How to Build AI Agents Your Business Can Trust (2027 Guide)](https://agnt.gg/articles/engineering/ai-harness-engineering): AI harness engineering for teams: the permissions, memory, evals and cost limits that make AI agents safe on business data. Sourced guide with a tested starter.
- [The Agent Loop, Explained: Plan, Act, Observe, Evaluate](https://agnt.gg/articles/agents/agent-loop-explained): An AI agent isn't one clever answer. It's a loop: plan, act, observe, evaluate. Watch the agent loop run as an interactive machine you can step through.
- [AI Harness vs Agent Framework vs Workflow: What Do You Actually Need?](https://agnt.gg/articles/comparisons/ai-harness-vs-agent-framework): Model, agent, harness, framework and workflow defined against one example. Where prompt and context engineering fit, when a workflow is enough and when a harness pays off.
- [Local vs Cloud vs Hybrid AI Assistants: Where Does Your Data Go?](https://agnt.gg/articles/comparisons/local-vs-cloud-ai-assistant): Compare local, cloud and hybrid AI assistants by tracing every component: interface, model, documents, tools, logs and backups. Retention, training, cost and a decision tree.
- [Private AI Assistants: How to Compare Data, Control and Exit Options](https://agnt.gg/articles/comparisons/private-ai-assistants): How to compare private AI assistants on evidence, not labels: data destinations, retention, permissions, inspectable memory, recovery and exit, with a hands-on acceptance test.
- [RAG vs Memory vs Context: Three Different Jobs in Your AI Assistant](https://agnt.gg/articles/comparisons/rag-vs-memory-vs-context): Context, retrieval (RAG) and memory solve different problems. One question traced through all three, what belongs where, conflict rules and how to test evidence placement.
- [AI Agent Logging: Understand Actions Without Leaking Secrets](https://agnt.gg/articles/engineering/ai-agent-logging): Design AI agent logs and run receipts that explain what ran, what changed and why it stopped, while keeping secrets and personal data out. Schema, redaction and retention.
- [AI Agent Permissions: Read, Draft, Approve, Execute](https://agnt.gg/articles/engineering/ai-agent-permissions): Design AI agent permissions enforced in code, not prompts: action classes, exact-action approval, expiry, one-time use, revocation and the negative tests that prove it.
- [Why Your AI Agent Loops, Forgets or Claims It Is Finished](https://agnt.gg/articles/engineering/ai-agent-troubleshooting): Diagnose AI agent failures by symptom: loops, forgetting, stale facts, false completion, duplicate actions and runaway cost. Find the responsible layer, fix it and add a regression test.
- [What Does a Personal AI Assistant Cost? A Reproducible Worksheet](https://agnt.gg/articles/engineering/personal-ai-assistant-cost): Work out what a personal AI assistant really costs per successful task: tokens, tools, hosting, hardware, energy and your time, with formulas, scenarios and budget limits.
- [How to Test a Personal AI Assistant Before Trusting It](https://agnt.gg/articles/engineering/personal-ai-assistant-evaluation): Evaluate a personal AI assistant on outcomes, not fluency: a 30-case suite, final-state checks, the right graders, pass@k vs pass^k, honest reporting and regression tests.
- [Personal AI Memory: Store, Inspect, Correct, Export and Delete](https://agnt.gg/articles/engineering/personal-ai-memory): Design personal AI memory you can see and change: what to remember, a record format with provenance, correction you can verify, portable export and honest deletion limits.
- [Prompt Injection in Personal Assistants: A Safe Testing Lab](https://agnt.gg/articles/engineering/personal-ai-prompt-injection): A defensive lab for prompt injection in personal AI assistants: synthetic fixtures, layered controls, scoring security and utility separately, and what passing does not prove.
- [Scheduling AI Workflows Without Duplicate Actions](https://agnt.gg/articles/engineering/schedule-ai-workflows-safely): Run scheduled AI workflows that never act twice: separate schedules from runs, checkpoint state, use idempotency keys, classify retries, reconcile uncertain outcomes and test failures.
- [Build a Meeting-Prep Assistant with Read-Only Calendar Access](https://agnt.gg/articles/tutorials/ai-meeting-prep-assistant): Build an AI meeting-prep assistant that writes sourced briefs from your calendar and notes without changing anything: scoped access, timezone tests and no-write checks.
- [How to Build Your Own AI Assistant: A Read-Only First Project](https://agnt.gg/articles/tutorials/build-your-own-ai-assistant): Build a first personal AI assistant with one read-only tool, a hard step limit and a run receipt. Tested 50-line Node.js starter, boundary tests and next steps.
- [Build an Email Assistant That Drafts but Does Not Send](https://agnt.gg/articles/tutorials/draft-only-ai-email-assistant): Build a personal AI email assistant that writes grounded drafts but has no ability to send: synthetic inbox, scoped reads, data flow, no-send proof and hostile-message tests.
- [Which Local AI Model Fits Your Hardware and Tasks?](https://agnt.gg/articles/tutorials/local-ai-assistant-hardware): Choose local AI hardware from your tasks, not parameter counts: memory budgets, quantisation, context and concurrency, plus a reproducible benchmark to run before buying anything.
- [MCP for Personal AI: Connect Tools Without Giving Away the Keys](https://agnt.gg/articles/tutorials/mcp-personal-ai-assistant): Connect a personal AI assistant to tools through MCP with scoped access: transports, credentials, token passthrough, tool contracts, denial tests and revocation.
- [Backup, Restore and Upgrade Your Personal AI Harness](https://agnt.gg/articles/tutorials/personal-ai-backup-restore): Back up and restore a personal AI harness properly: inventory state, write a manifest, restore into a clean environment, re-authorise deliberately and verify data and behaviour.
- [Personal Knowledge-Base AI with Verifiable Sources](https://agnt.gg/articles/tutorials/personal-knowledge-base-ai): Build a personal knowledge-base AI whose citations actually support its answers: scoped documents, provenance, retrieval checks, claim-to-source verification and stale-data tests.
- [Switch AI Providers Without Losing Your Workflows: A Practical Exit Test](https://agnt.gg/articles/tutorials/switch-ai-providers): Avoid AI vendor lock-in with a tested exit: inventory portable state, export it, adapt model and tool interfaces, rerun the same tasks, revoke old access and plan rollback.
- [$0.99 per outcome: notes on AI agent business ideas](https://agnt.gg/articles/agents/ai-agent-business-ideas): AI agent business ideas for 2026, sorted by how they get paid, with verified adoption, token-cost and cancellation data, and a 20-job test before you price.
- [AI Agents for Small Business: 10 Bottlenecks They Remove (and 3 They Don't)](https://agnt.gg/articles/agents/ai-agents-for-small-business): A practical guide to AI agents for small business: ten everyday bottlenecks agents can take off an owner's plate, what each costs to run, where to keep a human, and what adoption data says.
- [How AI Agents Work: The Loop, the Tools and the Harness Around Them](https://agnt.gg/articles/agents/how-ai-agents-work): How AI agents actually work, step by step: the model, the tool loop, memory, stopping conditions and the harness that makes it safe, with sources from Anthropic, OpenAI and the ReAct paper.
- [Zapier vs n8n vs AI Agents: Which Automation Approach Wins in 2026?](https://agnt.gg/articles/comparisons/zapier-vs-n8n-vs-ai-agents): Zapier vs n8n vs AI agents compared on pricing model, hosting, licence, AI features and control, with current prices and a worked cost example, plus when an agent beats both.
- [AI Agent Observability: See Every Trace, Tool Call and Cost](https://agnt.gg/articles/engineering/ai-agent-observability): AI agent observability explained: what to trace, the OpenTelemetry GenAI conventions, the metrics that matter, evals vs tracing, privacy, and how teams debug agents in production.
- [AI Agents vs Workflows: When to Use Each (With Examples)](https://agnt.gg/articles/engineering/ai-agents-vs-workflows): AI agent vs workflow, explained with Anthropic's definitions: when a fixed workflow beats an agent, when an agent is worth the cost, the five workflow patterns, and how to combine both.
- [Single-Agent vs Multi-Agent Systems: When One Agent Is Enough](https://agnt.gg/articles/engineering/single-agent-vs-multi-agent): Single-agent vs multi-agent AI systems compared with evidence from Anthropic, Cognition and OpenAI: when multiple agents win, what they cost in tokens, and the patterns that actually work.
- [The Rise of Local AI: Why Builders Are Running Models and Agents on Their Own Machines](https://agnt.gg/articles/industry/rise-of-local-ai): Why local AI took off: inference costs fell 280-fold, open-weight models closed the gap, and desktop hardware can now hold 100B+ parameter models. The trade-offs, and when local wins.
- [AI Agent Memory Explained: Types, Architectures and What Actually Works](https://agnt.gg/articles/research/ai-agent-memory): AI agent memory explained: working, episodic, semantic and procedural memory, the main architectures (MemGPT, Generative Agents, Mem0), why bigger context windows are not memory, and how to design it.
- [How to Build an AI Workflow: A Step-by-Step Guide for Your First Automation](https://agnt.gg/articles/tutorials/how-to-build-an-ai-workflow): Build your first AI workflow step by step: pick the job, draw the steps, choose the trigger, place the AI step, add a human gate, test, and run it on a schedule.
- [The 12 Best AI Agent Platforms in 2026](https://agnt.gg/articles/agents/best-ai-agent-platforms): The best AI agent platforms in 2026 for building and running agents: AGNT, n8n, Dust, Lindy, Gumloop, Relevance AI, Microsoft Copilot Studio, Zapier, Make, Dify and more.
- [The 12 Best Free AI Agents in 2026 (Actually Free)](https://agnt.gg/articles/agents/best-free-ai-agents): The best free AI agents in 2026, and what "free" really means for each: open-source licence, free tier limits, and the model bill you still pay. Honest picks with sources.
- [The 14 Best MCP Clients in 2026](https://agnt.gg/articles/agents/best-mcp-clients): The best MCP clients in 2026 compared: Claude, ChatGPT, VS Code, Cursor, Codex, Gemini CLI, LM Studio, Goose, Cline, AGNT and more, by transport, local-model support and cost.
- [The 11 Best Ollama GUIs in 2026 (Web, Desktop and Mobile)](https://agnt.gg/articles/agents/best-ollama-gui): The best Ollama GUIs and front ends in 2026: Open WebUI, AnythingLLM, LibreChat, Lobe Chat, Msty, Cherry Studio, Chatbox, Enchanted, Page Assist, AGNT and more, compared.
- [The 10 Best OpenClaw Alternatives in 2026](https://agnt.gg/articles/agents/best-openclaw-alternatives): The best OpenClaw alternatives in 2026, sorted by why you are leaving: security isolation, a lighter footprint, zero setup, a visual workflow canvas or team features. With sources.
- [The 11 Best Personal AI Agents in 2026](https://agnt.gg/articles/agents/best-personal-ai-agents): The best personal AI agents in 2026 compared honestly: OpenClaw, Hermes Agent, Meta Muse, Vellum, Claude, Lindy, AGNT and more, with where each runs and what it costs.
- [System 1 vs System 2 in AI: Fast by Default, Slow on Purpose](https://agnt.gg/articles/research/system-1-vs-system-2-ai): Why Jev, GLiNER, classifiers and reasoning LLMs belong in the same architecture, and how to decide when an AI system should think harder.
- [Model vs. Harness: Where Super Intelligence Actually Does Work](https://agnt.gg/articles/engineering/model-vs-harness): The model thinks. The harness works. A plain-language breakdown of what a language model does, what an agent harness does, how one real task splits between them, and why the harness decides whether super intelligence ships results or just talks.
- [Super Intelligence Needs a Harness](https://agnt.gg/articles/industry/super-intelligence-needs-a-harness): AI just got renamed "super intelligence." Whatever you call it, raw intelligence is only useful when something points it at real work, remembers what happened, asks before it acts and keeps a record. That something is the harness, and it is about to become the most important layer in the stack.
- [AI Is Now "Super Intelligence": What the SI Rename Actually Means](https://agnt.gg/articles/news/ai-renamed-super-intelligence): At the UN on September 22, 2026, President Trump said the United States will call artificial intelligence "super intelligence," or SI, in all official documents. Here is exactly what he said, where the word comes from, what changes, and what it means if you build with AI.
- [100 AI Automation Ideas You Can Build in 2026](https://agnt.gg/articles/agents/100-ai-automation-ideas): 100 practical AI automation ideas with triggers, actions, outputs, human checkpoints, build components, effort, and risk.
- [The 100 Best AI Agent Tools in 2026](https://agnt.gg/articles/agents/100-best-ai-agent-tools): 100 AI agent tools grouped by capability, with each tool’s function, agent use case, constraint, and direct first-party source.
- [The 100 Best Claude Code Skills in 2026](https://agnt.gg/articles/agents/100-best-claude-code-skills): 100 Claude Code skills and skill-bearing plugins, verified against Anthropic docs and public repos, with install and security notes.
- [The 100 Best MCP Servers in 2026: The Definitive Guide](https://agnt.gg/articles/agents/100-best-mcp-servers): 100 MCP servers grouped by job, with direct source links, selection advice, setup examples, and security notes for AI agents.
- [The 100 Best AI Agent Prompts and Templates in 2026](https://agnt.gg/articles/agents/ai-agent-prompts): 100 copyable AI agent prompt templates with required inputs, output contracts, tool rules, stop conditions, and failure checks.
- [75 Real-World AI Agent Use Cases in 2026](https://agnt.gg/articles/agents/ai-agent-use-cases): 75 AI agent use cases with autonomy loops, tools, human controls, failure modes, risk tiers, and cited deployed examples.
- [The 50 Best AI Agent Frameworks in 2026: Compared](https://agnt.gg/articles/agents/best-ai-agent-frameworks): 50 AI agent frameworks compared by language, orchestration model, best fit, limits, deployment style, and primary documentation.
- [The 50 Best Local AI Agents in 2026](https://agnt.gg/articles/agents/best-local-ai-agents): 50 local and self-hosted AI agents compared by execution, model, storage, telemetry, platform, license, and practical limits.
- [The 50 Best Open-Source and Source-Available AI Agents in 2026](https://agnt.gg/articles/agents/best-open-source-ai-agents): 50 open-source and source-available AI agents compared by license, deployment, capabilities, fit, limits, and source access.
- [How to Build an AI Agent in 2026: Complete Step-by-Step Guide](https://agnt.gg/articles/tutorials/how-to-build-an-ai-agent): Build one bounded research agent three ways, with tool schemas, budgets, approvals, retries, evaluations, tests, and security controls.
- [Connect Claude Code to AGNT and use your own subscription](https://agnt.gg/articles/tutorials/connect-claude-code): Sign in with the Claude plan you already pay for and run Claude models inside AGNT. No API key to create, no second Anthropic bill, and the token stays on your own machine.
- [Everything as Code: The Quiet Reformation of Creative Work](https://agnt.gg/articles/industry/everything-as-code): AI's real shift is not that machines can make artifacts. It is that artifacts are becoming editable, executable systems. The next platform is not the renderer. It is the operating layer around the renderer.
- [Skill Memory Is Not Skill Evolution: Hermes Agent vs AGNT SES](https://agnt.gg/articles/comparisons/hermes-ses-vs-agnt): Hermes Agent gives an autonomous worker a notebook of reusable procedures. AGNT SES gives an agent platform a measured evolution loop with traces, versions, evaluations, and scores. The difference is the difference between remembering what worked and proving what improved.
- [The 100 Best AI Agent Skills in 2026: The Definitive Guide](https://agnt.gg/articles/agents/100-best-ai-agent-skills): A comprehensive, research-backed guide to the 100 most powerful AI agent skills across 12 categories — from coding and design to security, finance, and full-stack automation. Curated from 5,200+ community skills, GitHub repositories, enterprise deployments, and expert sources.
- [The AI Briefing: Week of April 7–13, 2026](https://agnt.gg/articles/news/ai-briefing-2026-04-13): Anthropic hit $30B ARR and overtook OpenAI. Claude Mythos was deemed too dangerous to release. Sam Altman's home was firebombed. Meta launched Muse Spark. Google dropped Gemma 4. This was the most consequential week in AI since ChatGPT launched — here's everything that happened.
- [AGNT Memory — A Whitepaper on Self-Improving Agent Infrastructure](https://agnt.gg/articles/research/agnt-memory-whitepaper): How AGNT's evolution system observes every execution, extracts typed insights, and rewrites its own agents, workflows, and skills to compound knowledge over time
- [AGNT vs Hermes Agent: A Technical Teardown of Two Self-Improving Agent Systems](https://agnt.gg/articles/comparisons/agnt-vs-hermes-agent): A deep comparison of Nous Research's Hermes Agent and AGNT — from architecture and memory to skills, workflows, evolution loops, and the AGI-style iteration engine that sets AGNT apart.
- [The NN-Shaped Hammer: When to Learn and When to Just Write the Damn Code](https://agnt.gg/articles/engineering/the-nn-shaped-hammer): Every learned parameter is a liability. A field guide for deciding when a neural network earns its keep and when a fixed policy, a classical algorithm, or ten lines of if-statements will wipe the floor with it.
- [From Naive to Near-Optimal: An autonomous iterative approach to hybrid string compression](https://agnt.gg/articles/research/hybrid-string-compression): Evolving a naive Run-Length Encoder through 10 generations of machine-guided optimization to achieve an 80.9% improvement in combined performance
- [OpenClaw vs. Hermes vs. AGNT: AI Agent Architectures Compared](https://agnt.gg/articles/comparisons/openclaw-vs-hermes-vs-agnt): A source-code-level technical comparison of OpenClaw (the unified messaging gateway), Hermes (the self-improving agent by Nous Research), and AGNT — the full-stack AI command center with visual workflows, multi-agent orchestration, and autonomous goal execution.
- [How to Connect OpenAI Codex CLI to AGNT](https://agnt.gg/articles/tutorials/connect-openai-codex-cli): Step-by-step guide to connecting OpenAI Codex CLI to AGNT using device authentication, with troubleshooting tips and supported models
- [Karpathy's Autoresearch vs. AGNT: Autonomous Skill Evolution Compared](https://agnt.gg/articles/comparisons/karpathy-autoresearch-vs-agnt): A technical comparison of Andrej Karpathy's Autoresearch — a 630-line autonomous ML research loop — against AGNT's SkillForge architecture for autonomous skill evolution across multi-agent systems.
- [Executor vs. AGNT: AI Agent Tool Infrastructure Compared](https://agnt.gg/articles/comparisons/executor-vs-agnt): A deep-dive technical comparison of Rhys Sullivan's Executor — a local-first sandboxed tool runtime for AI agents — against the AGNT platform's integrated tool, MCP, and workflow ecosystem.
- [Pi-AI Provider Migration: Decision Report — Why We're Not Migrating](https://agnt.gg/articles/engineering/pi-migration-report): A thorough technical evaluation of migrating AGNT's LLM provider system to @mariozechner/pi-ai. After quantifying maintenance costs, migration effort, ROI, and dependency risks, we recommend against migration and propose targeted improvements instead.
- [The Complete Guide to AI Agent Architectures in 2026](https://agnt.gg/articles/agents/ai-agent-architectures-guide): From single-agent loops to multi-agent orchestration — a deep dive into every architecture pattern powering today's most capable AI systems.
- [AI Agent Evaluation & Gold Standard System Plan](https://agnt.gg/articles/research/agent-eval-rl-system): A comprehensive evaluation framework for AI agent platforms using gold standard datasets, trace-based scoring, and LLM-as-a-Judge evaluators.
- [Building Autonomous Agents with the AGNT API](https://agnt.gg/articles/agents/building-autonomous-agents): A technical guide to designing and deploying autonomous AI agents using the AGNT API, covering endpoint integration, architecture patterns, and real-world implementation.
- [A Hippocampus Inspired Autonomous Learning System](https://agnt.gg/articles/research/hippocampus-learning-system): A biologically inspired cognitive memory framework modeled after the hippocampal-neocortical system for long-term reasoning and autonomous learning.
- [Recursive Self-Improvement in Autonomous Intelligence Systems](https://agnt.gg/articles/research/recursive-self-improvement): A Unified Architecture for Context, Memory, Reasoning, and Recursive Self-Improvement in Autonomous Intelligence Systems
- [Zero Latency AI - Bidirectional Streaming with Claude Agent SDK](https://agnt.gg/articles/tutorials/claude-agent-sdk-streaming): How to achieve instant AI responses using persistent socket connections, multi-agent architectures, and the agent-sdk's bidirectional streaming—eliminating cold start delays entirely
- [Claude Agent SDK Cheatsheet](https://agnt.gg/articles/tutorials/claude-agent-sdk-cheatsheet): Complete TypeScript/JavaScript reference for the Claude Agent SDK including installation, core functions, configuration options, tools, hooks, and migration guide

## Documentation

- [API Conventions](https://agnt.gg/docs/api/conventions): Errors, rate limiting, pagination, file uploads, SSE and WebSockets — shared behaviour across all endpoints.
- [API Quickstart](https://agnt.gg/docs/api): Base URLs, authentication in ten lines, and your first call to the AGNT API.
- [Authentication](https://agnt.gg/docs/api/authentication): The JWT token, the three places it lives, and how each context authenticates.
- [Getting Started with AGNT](https://agnt.gg/docs/getting-started): Install AGNT, connect a model, and give an agent its first real job — in about ten minutes.
- [Run AGNT with Docker](https://agnt.gg/docs/docker): Headless AGNT on a server or VPS: the full image and the lite image, from the product quickstarts.
- [Agents & Orchestration](https://agnt.gg/docs/api/agents): Create, update, chat with and stream from agents; orchestrator runs, async tool executions and agent groups.
- [Content & Artifacts](https://agnt.gg/docs/api/content-artifacts): AGNT local API reference for generated content outputs, artifacts, images and local files produced by agents and workflows.
- [Execution & Streaming](https://agnt.gg/docs/api/execution-streaming): AGNT local API reference for execution history, live execution streams and speech: inspect past runs and follow live ones.
- [Experiments & Contracts](https://agnt.gg/docs/api/experiments-contracts): AGNT local API reference for A/B experiments, behavioural contracts and mutation history: measure and govern how agents change.
- [FileSystem](https://agnt.gg/docs/api/filesystem): AGNT local API reference for the sandboxed file operations available to agents and workflows: read, write and list files.
- [Goals & Autonomy](https://agnt.gg/docs/api/goals-autonomy): Goal creation and autonomous execution, plus the closed-loop safety system that governs unattended runs.
- [Ledger & Wallets](https://agnt.gg/docs/api/ledger-wallets): AGNT local API reference for the cost ledger behind every run receipt, and for wallet balances, so spend is visible per run.
- [Memory & Insights](https://agnt.gg/docs/api/memory-insights): Persistent agent memory, and the evolution/insight system that learns from runs.
- [Models & Providers](https://agnt.gg/docs/api/models-providers): Model catalogues, provider connections, OAuth flows and custom providers.
- [Phone Access & Pairing](https://agnt.gg/docs/api/pairing): Pair a phone by QR code, the security model behind it, and the LAN access toggle.
- [Plugins & MCP](https://agnt.gg/docs/api/plugins-mcp): AGNT local API reference to install and manage plugins, and to connect MCP servers as tool sources for agents and workflows.
- [Schedules & Listeners](https://agnt.gg/docs/api/schedules): Cron-style schedules and email listeners that start work without a human.
- [Skills](https://agnt.gg/docs/api/skills): AGNT local API reference for skill packages, SkillForge authoring and skill discovery, including SKILL.md import and export.
- [Tools](https://agnt.gg/docs/api/tools): AGNT local API reference for built-in tools, custom tools, tool schemas and npm dependencies: everything an agent can call.
- [Users & System](https://agnt.gg/docs/api/system): AGNT local API reference for local user state, system information and admin endpoints on your own AGNT instance.
- [Workflows](https://agnt.gg/docs/api/workflows): AGNT local API reference for workflows: create, run, import and version them, and start them from webhook triggers.
- [Workspace & UI](https://agnt.gg/docs/api/workspace-ui): AGNT local API reference for layouts, workspaces, widget definitions and conversation settings in the desktop app.

## Whitepapers

- [A Proof of Graffiti 290 — AGNT Labs Research Report](https://agnt.gg/whitepapers/a-proof-of-graffiti-290): A complete proof of Graffiti 290, machine-checked in Lean 4: minus the second-smallest adjacency eigenvalue of a girth-five graph is at most its size over mean gravity.
- [A Proof of Graffiti 292 — AGNT Labs Research Report](https://agnt.gg/whitepapers/a-proof-of-graffiti-292): A complete proof of Graffiti 292, machine-checked in Lean 4: the least positive adjacency eigenvalue of a girth-five graph is at most n over mean gravity.
- [A Proof of Graffiti 295 — AGNT Labs Research Report](https://agnt.gg/whitepapers/a-proof-of-graffiti-295): A complete proof of Graffiti 295, machine-checked in Lean 4: a girth-five graph has at most n over mean gravity positive distance eigenvalues.
- [The Anatomy of the First Jacobian Counterexample — AGNT Labs Technical Report II (v1.0)](https://agnt.gg/whitepapers/anatomy-of-the-first-jacobian-counterexample): A structural analysis of Alpöge's Jacobian Conjecture counterexample: fold and torus theorems in exact arithmetic, with hashed replayable certificates.
- [Cyclic Rigidity in Dimension Two — AGNT Labs Technical Report IV](https://agnt.gg/whitepapers/cyclic-rigidity-jacobian-conjecture-dimension-two): A Newton-edge proof of cyclic rigidity for two-variable Keller maps, with exact symbolic checks, independent implementation, adversarial search, and reproducibility artifacts.
- [Erdős Problems 552 and 85 — AGNT Labs Research Report](https://agnt.gg/whitepapers/erdos-552-and-85-chens-bound-and-an-equivalence): A Lean 4 proof of Chen's bound for Erdős Problem 552, an equivalence showing Erdős Problem 85 is a question about plateaus of R(C4, K1,n), and three corrections.
- [Erdős 835: A Primality Sieve and Machine-Checked Small Cases — AGNT Labs](https://agnt.gg/whitepapers/erdos-835-a-primality-sieve-and-machine-checked-small-cases): A verified reduction of Erdos Problem 835 to the chromatic number of the middle Johnson graph J(2k,k), an exact packing-bound argument identifying YES-instances with large sets of Steiner systems S(k-1,k,2k), a machine-verified primality sieve (design divisibility holds iff k+1 is prime) that independently reproves the Ma-Tang composite direction, and five-oracle machine-checked settlement of the small cases k=2,3,4,5.
- [Graffiti 284 Refuted by Hoffman–Singleton — AGNT Labs Verification Note](https://agnt.gg/whitepapers/graffiti-284-refutation): Independent exact verification that the Hoffman–Singleton graph refutes Graffiti 284: the minimum dual degree exceeds minus the smallest distance-matrix eigenvalue. Priority credited to the Capy Build agent run of 2026-07-22.
- [Machine-Verified Corollary Mining of the Jacobian Conjecture Collapse (v1.0)](https://agnt.gg/whitepapers/machine-verified-corollary-mining-jacobian-conjecture): Two consequences of the Jacobian Conjecture's collapse, executed and machine-verified in exact rational arithmetic within a day of the announcement.
- [Power-Weighted Lifts — AGNT Labs Technical Report III](https://agnt.gg/whitepapers/power-weighted-lifts-higher-weight-keller-maps): A two-parameter power-weighted construction of explicit higher-weight noninjective Keller maps in three variables, with exact proofs, certificates, charts, and independent reproduction.
- [The Symmetric-Product Origin of the Jacobian Counterexample — AGNT Labs Technical Report IV (v1.0)](https://agnt.gg/whitepapers/sym3-origin-of-the-jacobian-counterexample): Where the Jacobian counterexample comes from: a three-to-one symmetric-product insert map, reduced to an explicit non-injective self-map of affine 3-space.
- [The Cache Wars 2: An Empirical Study of Prompt-Cache Efficiency in Six LLM Agent Harnesses](https://agnt.gg/whitepapers/the-cache-wars-2): A controlled measurement of prompt-cache efficiency across six production LLM agent harnesses on subscription-authenticated Claude and Codex paths, with a parametric cost model and complete artifacts.
- [An Empirical Study of Prompt-Cache Efficiency in Six LLM Agent Harnesses (v2.0)](https://agnt.gg/whitepapers/the-cache-wars-prompt-cache-efficiency-llm-agent-harnesses): A controlled measurement of prompt-cache efficiency across six production LLM agent harnesses, using byte-identical workloads and a 390-second pause test.
- [Trophies Are Not Tools: On Priming LLM Agents with Records of Their Own Past Success (v1.0)](https://agnt.gg/whitepapers/trophies-are-not-tools-priming-agents-with-past-success): Does priming an agent with records of its own past successes help? Measured against the published record and deployed telemetry: the intuition does not survive.

## Comparisons

- [Best Agentic Operating Systems & AI Agents (2026)](https://agnt.gg/compare/agentic-operating-systems): AGNT, Claude Code, OpenAI Codex, Hermes Agent, OpenClaw, Meta Muse, Vellum and Claude Cowork compared: what each agentic system is for, where it runs, what it costs.
- [AGNT vs ChatGPT AgentKit — Provider-Neutral vs Single-Vendor](https://agnt.gg/compare/agnt-vs-chatgpt-agentkit): AGNT vs ChatGPT AgentKit: provider-neutral local runtime versus single-vendor cloud agents. Where OpenAI's ecosystem wins, and where it costs you.
- [AGNT vs Claude Code: Agent OS vs Coding Agent (2026)](https://agnt.gg/compare/agnt-vs-claude-code): AGNT vs Claude Code, compared honestly: where Anthropic’s coding agent wins, where an agent operating system fits, and how to run Claude Code inside AGNT.
- [AGNT vs Claude Cowork: Agent OS vs Claude Tasks (2026)](https://agnt.gg/compare/agnt-vs-claude-cowork): AGNT vs Claude Cowork: Anthropic’s agentic tasks for knowledge work, now merging into Claude, against a local-first agent OS. Where each one fits.
- [AGNT vs OpenAI Codex: Agent OS vs Coding Agent (2026)](https://agnt.gg/compare/agnt-vs-codex): AGNT vs OpenAI Codex, compared honestly: the CLI, app and cloud agent against a local-first agent OS, where each wins, and how to delegate to Codex from AGNT.
- [AGNT vs CrewAI — which agent stack fits?](https://agnt.gg/compare/agnt-vs-crewai): CrewAI and AGNT solve different problems. CrewAI orchestrates role-based agent crews in Python. AGNT gives multi-agent work a full runtime: visual orchestration, memory, a tool vault, receipts, and a marketplace of reusable parts.
- [AGNT vs Hermes Agent: Agent OS vs Self-Improving Agent](https://agnt.gg/compare/agnt-vs-hermes-agent): AGNT vs Hermes Agent by Nous Research: an MIT-licensed self-improving agent that lives in your chat apps, against a local-first agent OS. Where each one wins.
- [AGNT vs LangChain — which agent stack fits?](https://agnt.gg/compare/agnt-vs-langchain): LangChain and AGNT solve different problems. LangChain is a developer framework — a box of parts for building agent systems in code. AGNT is a running product: visual workflows, a marketplace, receipts and a local runtime, usable the hour you install it.
- [AGNT vs Make — Operations Pricing vs Local Execution](https://agnt.gg/compare/agnt-vs-make): AGNT vs Make: operations-based cloud pricing versus unlimited local execution. Where Make's visual scenario builder wins, and where owning the runtime does.
- [AGNT vs Meta Muse: Agent OS vs Personal AI Agent](https://agnt.gg/compare/agnt-vs-meta-muse): AGNT vs Meta Muse, Meta’s personal AI agent: a cloud VM agent in WhatsApp against a local-first agent OS you run yourself. Who each one is for, honestly.
- [AGNT vs n8n — Agent Runtime vs Workflow Automation](https://agnt.gg/compare/agnt-vs-n8n): AGNT vs n8n: both self-host, but the primitive differs. n8n's unit is the workflow; AGNT's is the agent with memory, goals and receipts. An honest comparison.
- [AGNT vs OpenClaw: Agent OS vs Personal AI Assistant](https://agnt.gg/compare/agnt-vs-openclaw): AGNT vs OpenClaw: the MIT-licensed assistant that lives in 20+ chat apps, against a local-first agent operating system. An honest, row-by-row comparison.
- [AGNT vs Vellum: Agent OS vs Personal AI Assistant (2026)](https://agnt.gg/compare/agnt-vs-vellum): AGNT vs Vellum’s open-source personal AI assistant: long-term memory and a custom personality against a local-first agent OS with workflows and receipts.
- [AGNT vs Zapier — Local Runtime vs Metered Cloud](https://agnt.gg/compare/agnt-vs-zapier): AGNT vs Zapier: an owned local runtime with unmetered execution versus a cloud platform billed per task. Where each one wins, stated plainly.
- [Best Make.com Alternatives for AI Automation (2026)](https://agnt.gg/alternatives/make): The best Make.com alternatives in 2026, compared honestly: AGNT, n8n, Zapier, Activepieces and Gumloop, plus AI agents, with current prices, sources and trade-offs.
- [Best n8n Alternatives for AI Agents (2026)](https://agnt.gg/alternatives/n8n): The best n8n alternatives for AI agents in 2026, compared honestly: AGNT, Make, Zapier, Activepieces, Gumloop, Hermes Agent, OpenClaw and more, with prices and sources.
- [Best Zapier Alternatives for AI Automation (2026)](https://agnt.gg/alternatives/zapier): Zapier alternatives in 2026, compared honestly: AGNT, n8n, Make, Activepieces, Gumloop and AI agents, with prices and when you're better off staying.

## Product and policy

- [AGNT — Free local-first AI agent platform for Windows, macOS & Linux](https://agnt.gg/): AGNT is a free, local-first AI agent platform for Windows, macOS and Linux. Run agents, workflows and coding agents with your AI subscriptions or local models.
- [Pricing & Plans](https://agnt.gg/pricing): AGNT Pricing & Plans: Free Community Core for local execution, Business Cloud for teams, and Managed Operations for enterprise scale.
- [Download Community Core](https://agnt.gg/downloads/): Download AGNT Community Core free for Windows, macOS, and Linux. Local-first AI operations desktop app with verified receipts.
- [Product — The Local-First Agent Runtime](https://agnt.gg/product/): AGNT is a local-first agent runtime: agents with memory, visual workflows, autonomous goals, MCP tools, and verifiable receipts — running on your hardware.
- [AI Agents with Memory and Tools](https://agnt.gg/product/agents): AGNT agents are persistent AI workers with their own memory, tools, model routing and budgets — they remember prior runs instead of starting cold.
- [Visual AI Workflow Builder](https://agnt.gg/product/workflows): A visual AI workflow builder with 119+ tool nodes, triggers, branching and version rollback — deterministic where it matters, agentic where it helps.
- [Autonomous AI Goals — Plan, Execute, Evaluate, Replan](https://agnt.gg/product/goals): Autonomous AI goals: describe an outcome and AGNT plans, executes, evaluates and replans until your success criteria are met — or reports that they weren't.
- [Agent memory](https://agnt.gg/product/memory): AGNT agents carry persistent memory across sessions — facts, preferences, corrections and learned patterns — so the tenth run is sharper than the first.
- [Skills](https://agnt.gg/product/skills): A skill is a versioned playbook an agent can load on demand — your process, standards and examples, applied consistently by every agent that carries it.
- [Plugins](https://agnt.gg/product/plugins): Plugins add new tools, providers and UI to AGNT — built from chat, installed with hot-reload, and shared through the marketplace.
- [MCP Client — Model Context Protocol Tools](https://agnt.gg/product/mcp): AGNT is a full Model Context Protocol client — point it at any MCP server and those tools become callable by every agent, alongside 119+ native nodes.
- [Local-First AI & Data Sovereignty](https://agnt.gg/product/local-first): Local-first AI and data sovereignty: files, credentials, memory and receipts stay on your machine. Zero telemetry, fully air-gappable with a local model.
- [Self-Hosted AI Automation](https://agnt.gg/product/self-hosted): Self-hosted AI automation: run AGNT on your own server or in Docker. No per-task metering, no execution caps, no vendor holding your automation hostage.
- [About AGNT](https://agnt.gg/about): AGNT exists because AI work should be verifiable. We build a local-first runtime where agents hold real jobs, use real tools, and produce receipts for everything they do.
- [Security & Privacy Architecture](https://agnt.gg/security): AGNT Security & Privacy Architecture: Zero telemetry, local credential vault encryption, approval gates, and verifiable audit receipts.
- [AGNT Marketplace — Ready-to-run AI work](https://agnt.gg/marketplace): Download ready-to-run AI agents, tools, workflows, skills, plugins, and widgets for AGNT. Start with a proven template and customize it in seconds.
- [Changelog & Releases](https://agnt.gg/changelog): AGNT changelog and release notes. Every version, every new feature, improvement and fix — searchable, with downloads for each release.
- [Nathan Wilbanks, founder of AGNT](https://agnt.gg/authors/nathan-wilbanks): Founder of AGNT, the local-first agent OS. Systems architect and AI, ML and RL researcher, and an exited founder, building AGNT from Los Altos, California.
- [Annie, AGNT’s AI agent](https://agnt.gg/authors/annie): Annie is the AI agent built into AGNT, the local-first agent OS. Articles credited to Annie are written by Annie and published by AGNT. Annie is software, not a person.

## Integrations

- [AI agents for Slack](https://agnt.gg/integrations/slack): Team chat where operations actually surface.
- [AI agents for Gmail](https://agnt.gg/integrations/gmail): The inbox most operations still start in.
- [AI agents for Google Calendar](https://agnt.gg/integrations/googlecalendar): The shared clock your workflows can read and write.
- [AI agents for Google Drive](https://agnt.gg/integrations/googledrive): Where the documents live — and pile up.
- [AI agents for Google Sheets](https://agnt.gg/integrations/googlesheets): The database every team already has open.
- [AI agents for Google Slides](https://agnt.gg/integrations/googleslides): Decks assembled from real data instead of Friday-night copy-paste.
- [AI agents for Notion](https://agnt.gg/integrations/notion): The team wiki agents can actually keep current.
- [AI agents for Airtable](https://agnt.gg/integrations/airtable): Structured records with an API worth automating.
- [AI agents for GitHub](https://agnt.gg/integrations/github): Where the code — and half the process — lives.
- [AI agents for Jira](https://agnt.gg/integrations/jira): The ticket system of record for engineering work.
- [AI agents for Linear](https://agnt.gg/integrations/linear): Fast issue tracking that deserves fast automation.
- [AI agents for Asana](https://agnt.gg/integrations/asana): Team tasks, due dates and the follow-through between them.
- [AI agents for Trello](https://agnt.gg/integrations/trello): Boards and cards — simple enough to automate deeply.
- [AI agents for Discord](https://agnt.gg/integrations/discord): Community chat with an automation-friendly API.
- [AI agents for Telegram](https://agnt.gg/integrations/telegram): The fastest channel between your agents and your phone.
- [AI agents for Twilio](https://agnt.gg/integrations/twilio): SMS and voice when the alert genuinely cannot wait.
- [AI agents for X (Twitter)](https://agnt.gg/integrations/x): The feed where your market thinks out loud.
- [AI agents for YouTube](https://agnt.gg/integrations/youtube): Long-form content worth repurposing everywhere else.
- [AI agents for HubSpot](https://agnt.gg/integrations/hubspot): The CRM your pipeline reporting depends on.
- [AI agents for Salesforce](https://agnt.gg/integrations/salesforce): The enterprise system of record for revenue.
- [AI agents for Stripe](https://agnt.gg/integrations/stripe): Payments, subscriptions and the events behind revenue.
- [AI agents for QuickBooks](https://agnt.gg/integrations/quickbooks): The books — kept clean without the keying.
- [AI agents for Calendly](https://agnt.gg/integrations/calendly): Bookings that should kick off preparation automatically.
- [AI agents for Dropbox](https://agnt.gg/integrations/dropbox): Shared files that arrive needing processing.
- [AI agents for Obsidian](https://agnt.gg/integrations/obsidian): A local-first vault — a natural fit for a local-first runtime.
- [AI agents for PostgreSQL](https://agnt.gg/integrations/postgresql): The production database your reports should come from.
- [AI agents for Redis](https://agnt.gg/integrations/redis): Fast state for coordination between workflow runs.
- [AI agents for Docker](https://agnt.gg/integrations/docker): Containers — including the one AGNT itself can run in.
- [AI agents for AWS](https://agnt.gg/integrations/aws): The infrastructure bill and the services behind it.
- [AI agents for Azure](https://agnt.gg/integrations/azure): Microsoft cloud, wired into your operational reporting.
- [AI agents for Microsoft 365](https://agnt.gg/integrations/microsoft): Outlook, Teams and Office — the other half of many stacks.
- [AI agents for Vercel](https://agnt.gg/integrations/vercel): Deployments your team should hear about in context.
- [AI agents for Netlify](https://agnt.gg/integrations/netlify): Sites and builds reporting into your workflows.
- [AI agents for Sentry](https://agnt.gg/integrations/sentry): Errors that deserve summaries, not just alerts.
- [AI agents for Firecrawl](https://agnt.gg/integrations/firecrawl): Clean page extraction for research-grade scraping.
- [AI agents for Unsplash](https://agnt.gg/integrations/unsplash): Licensed imagery for generated content.
- [AI agents for Weights & Biases](https://agnt.gg/integrations/wandb): Experiment tracking your reports can read.
- [AI agents for Zapier](https://agnt.gg/integrations/zapier): Keep your Zaps — give them a reasoning layer.
- [AI agents for NVIDIA](https://agnt.gg/integrations/nvidia): NIM endpoints and GPU-served models in your workflows.

## Use cases

- [Email triage, done before you open the inbox](https://agnt.gg/marketplace/email-triage): An AGNT agent reads incoming mail, classifies it, drafts replies for the routine cases, and escalates the ones that genuinely need you — all under rules you write.
- [Customer support that answers from your own docs](https://agnt.gg/marketplace/customer-support): Agents resolve the questions your docs already answer and hand the rest to your team with full context — history, account state, and a suggested reply.
- [A first pass on every pull request](https://agnt.gg/marketplace/code-review): An agent reviews diffs against your standards — bugs, security, style, missing tests — and leaves structured comments before a human ever looks.
- [Research briefings, sourced and cited](https://agnt.gg/marketplace/research-briefings): Point an agent at a question; get back a sourced briefing — market landscape, technical comparison, diligence pass — with every claim linked.
- [Competitor monitoring that reports what changed](https://agnt.gg/marketplace/competitor-monitoring): Agents watch competitor sites, changelogs, pricing pages and news — and brief you on what actually changed, with links to the evidence.
- [Content production, from idea to published](https://agnt.gg/marketplace/content-pipeline): Run research, drafting, review and publishing as one workflow: agents do the legwork, humans hold the gates that matter.
- [Social posts written native to each platform](https://agnt.gg/marketplace/social-media-automation): Turn published work into platform-native posts, schedule them, and collect the engagement numbers back into one report.
- [Release notes written from what actually merged](https://agnt.gg/marketplace/write-release-notes): An agent reads every pull request merged since your last release and writes the notes your team needs — what changed, what to watch, and what needs a migration.
- [Dependency advisories, sorted by what touches you](https://agnt.gg/marketplace/monitor-dependency-advisories): An agent reads the security updates Dependabot or Renovate opens, searches your code for the affected package, and tells you which advisories actually reach code you run.
- [Where is this handled? Answered with file paths](https://agnt.gg/marketplace/answer-codebase-questions): Ask how something works in your repository and get an answer that points at the files, functions and call sites involved — read from the code, not remembered.
- [Pull request descriptions a reviewer can start from](https://agnt.gg/marketplace/draft-pr-descriptions): An agent reads the diff, the commits and the linked issue, and writes the description your reviewer needs — what changed, why, how to test it and where to look hardest.
- [A changelog kept current, one pull request at a time](https://agnt.gg/marketplace/maintain-changelog): An agent adds an entry for every merged pull request to your changelog under Unreleased, proposes it as a pull request, and rolls it into a version when you cut a release.

## Solutions by team

- [AGNT for sales teams](https://agnt.gg/solutions/sales): Every lead enriched, every meeting logged, every follow-up drafted — the CRM keeps itself current while reps talk to buyers.
- [AGNT for marketing teams](https://agnt.gg/solutions/marketing): Research, drafting, repurposing and reporting — agents take the production work while your team keeps strategy and taste.
- [AGNT for engineering teams](https://agnt.gg/solutions/engineering): PR triage, release notes, incident summaries, dependency watch — agents handle the process work that surrounds shipping.
- [AGNT for operations teams](https://agnt.gg/solutions/operations): Data entry, reconciliation, vendor tracking, recurring reports — turned into verified workflows with receipts on every run.
- [AGNT for support teams](https://agnt.gg/solutions/support): Docs-grounded answers for the repetitive tickets, rich escalation for the hard ones, and a knowledge base that learns from resolutions.
- [AGNT for finance teams](https://agnt.gg/solutions/finance): Invoice processing, expense review, variance reporting — agents do the collection and matching, humans approve what posts.
- [AGNT for agencies](https://agnt.gg/solutions/agencies): Client reporting, research, content production and monitoring — productised as workflows you run for every client at once.
- [AGNT for founders](https://agnt.gg/solutions/founders): One person with AGNT covers research, outreach drafts, support and reporting — the operational load that usually eats the week.

## Model providers

- [OpenAI with AGNT](https://agnt.gg/providers/openai): GPT-class models for general reasoning, tool use and code.
- [Anthropic with AGNT](https://agnt.gg/providers/anthropic): Claude models — long context, careful reasoning, strong code.
- [Google Gemini with AGNT](https://agnt.gg/providers/gemini): Google’s multimodal family with massive context windows.
- [xAI Grok with AGNT](https://agnt.gg/providers/grok): xAI’s models with real-time X data affinity.
- [Meta Llama with AGNT](https://agnt.gg/providers/meta): The open-weight family the local ecosystem is built on.
- [Mistral with AGNT](https://agnt.gg/providers/mistral): European frontier lab with strong small models.
- [DeepSeek with AGNT](https://agnt.gg/providers/deepseek): Frontier-quality reasoning at disruptive cost.
- [Qwen with AGNT](https://agnt.gg/providers/qwen): Alibaba’s open family — dense to massive MoE.
- [Kimi with AGNT](https://agnt.gg/providers/kimi): Moonshot’s long-context agentic models.
- [MiniMax with AGNT](https://agnt.gg/providers/minimax): Efficient frontier models with huge context.
- [Z.ai with AGNT](https://agnt.gg/providers/zai): GLM models tuned for agentic work.
- [Cohere with AGNT](https://agnt.gg/providers/cohere): Enterprise-focused models with first-class RAG.
- [Groq with AGNT](https://agnt.gg/providers/groq): LPU-served open models at extreme speed.
- [Cerebras with AGNT](https://agnt.gg/providers/cerebras): Wafer-scale inference for open models.
- [Together AI with AGNT](https://agnt.gg/providers/together): One API over the open-weight universe.
- [OpenRouter with AGNT](https://agnt.gg/providers/openrouter): Every major model behind one key.
- [Replicate with AGNT](https://agnt.gg/providers/replicate): Run community models — vision, audio, image — by API.
- [Hugging Face with AGNT](https://agnt.gg/providers/huggingface): The hub — inference endpoints over the open ecosystem.
- [Perplexity with AGNT](https://agnt.gg/providers/perplexity): Answer models with live web grounding built in.
- [Ollama with AGNT](https://agnt.gg/providers/ollama): One command to run open models on your own machine.
- [LM Studio with AGNT](https://agnt.gg/providers/lmstudio): A desktop app for local models with an OpenAI-style server.
- [ElevenLabs with AGNT](https://agnt.gg/providers/elevenlabs): Production text-to-speech for agents that talk.

## Definitions

- [AI agent](https://agnt.gg/glossary/ai-agent): An AI system that pursues a goal by taking actions — calling tools, reading data, making decisions — rather than only generating text. Where a chatbot answers, an agent does: it plans steps, executes them against real systems, observes results and adjusts.
- [Agentic workflow](https://agnt.gg/glossary/agentic-workflow): A workflow where one or more steps are performed by a reasoning agent instead of fixed logic. Traditional automation branches on rules written in advance; an agentic workflow can read, judge and decide at the points where rules run out.
- [Autonomous agent](https://agnt.gg/glossary/autonomous-agent): An agent that operates without step-by-step supervision: given an objective, it plans, acts, evaluates its own results and continues until done or blocked. Autonomy is a spectrum — most production systems bound it with approval gates and budgets.
- [Multi-agent system](https://agnt.gg/glossary/multi-agent-system): Multiple agents cooperating on work too broad for one — a researcher feeding a writer feeding a reviewer, or specialists coordinated by an orchestrator. The hard problems are hand-off quality and shared state.
- [Agent orchestration](https://agnt.gg/glossary/agent-orchestration): Coordinating which agent or tool acts, in what order, with what context — routing tasks, merging results, retrying failures and enforcing gates. Orchestration turns capable-but-isolated models into a system.
- [Agent memory](https://agnt.gg/glossary/agent-memory): Persistent state an agent carries across sessions: facts about its user, corrections received, patterns that worked. Without it every run starts from zero; with it the tenth run outperforms the first.
- [Agent skills](https://agnt.gg/glossary/agent-skills): Packaged, versioned expertise an agent loads on demand — a playbook containing process, standards and examples for one class of task. Skills make competence reusable and reviewable instead of trapped in prompts.
- [MCP (Model Context Protocol)](https://agnt.gg/glossary/mcp): An open protocol for connecting AI models to tools and data sources through a standard interface. Instead of custom integrations per model per tool, MCP servers expose capabilities any compliant client can use.
- [MCP server](https://agnt.gg/glossary/mcp-server): A process exposing tools, resources or prompts over the Model Context Protocol. One well-built server — for a database, an API, a filesystem — becomes usable by every MCP-capable client without further glue.
- [Tool calling](https://agnt.gg/glossary/tool-calling): A model invoking external functions — search, database queries, file writes — by emitting structured calls its runtime executes. Tool calling is what turns language models from advisors into actors.
- [Function calling](https://agnt.gg/glossary/function-calling): The model-side mechanism behind tool use: the model returns a structured invocation — name plus JSON arguments — matching a declared schema, which the host executes and feeds back.
- [LLM runtime](https://agnt.gg/glossary/llm-runtime): The environment that turns model output into real work: executing tool calls, managing context and memory, enforcing permissions, retrying failures, recording what happened. Models reason; runtimes act.
- [Local-first AI](https://agnt.gg/glossary/local-first-ai): AI software whose data, credentials and execution live on hardware you control, with cloud as an option rather than a requirement. Privacy, cost and ownership follow from that architectural choice.
- [Self-hosted AI](https://agnt.gg/glossary/self-hosted-ai): Running AI infrastructure — runtimes, models, or both — on your own servers. Teams self-host for data control, compliance, predictable cost and independence from vendor availability.
- [Prompt caching](https://agnt.gg/glossary/prompt-caching): Reusing computation for the repeated prefix of prompts — system instructions, tool schemas, standing context — so consecutive calls pay only for what changed. For agent loops sharing a large stable prefix, cost and latency can drop dramatically.
- [RAG (Retrieval-Augmented Generation)](https://agnt.gg/glossary/rag): Fetching relevant documents at query time and placing them in the model’s context, so answers ground in your data instead of training memory. The retrieval layer determines most of the quality.
- [Context window](https://agnt.gg/glossary/context-window): The maximum text a model attends to in one call — instructions, history, retrieved material and its own output. Managing what earns a place in that window is a core engineering discipline of agent systems.
- [Fine-tuning](https://agnt.gg/glossary/fine-tuning): Continuing a model’s training on domain examples to shift behaviour — terminology, format, tone — into the weights. Powerful when in-context approaches plateau, at the price of maintenance per model version.
- [Inference](https://agnt.gg/glossary/inference): Running a trained model to produce output — the operational, per-request phase of AI, as opposed to training. Latency, throughput and cost per token are its economics.
- [Token](https://agnt.gg/glossary/token): The unit models read and write — roughly three-quarters of an English word. Context limits and API pricing are both denominated in tokens, which makes token discipline cost discipline.
- [System prompt](https://agnt.gg/glossary/system-prompt): The standing instruction that defines a model’s role, rules and boundaries before any user input arrives. In agent systems it carries identity, tool policy and safety rails — and deserves version control.
- [Webhook automation](https://agnt.gg/glossary/webhook-automation): Triggering workflows from HTTP callbacks the moment events happen — a payment, a form, a commit — instead of polling on a schedule. Webhooks are the event backbone of most integration work.
- [Human-in-the-loop](https://agnt.gg/glossary/human-in-the-loop): Automation that pauses for human judgment at defined points — approving an outbound email, releasing a payment, accepting a diff. The design question is which decisions warrant a human, not whether automation is safe.
- [Execution receipt](https://agnt.gg/glossary/execution-receipt): A structured record an agent produces for each operation: what was read, which tools ran with what arguments, what was decided and what was written. Receipts make AI work reviewable and attributable after the fact.
- [What is an AI agent?](https://agnt.gg/what-is/an-ai-agent): An AI agent is software that pursues a goal by acting — not just answering. Give it an objective and it plans steps, calls tools, reads results and keeps going until the job is done or a human gate stops it.
- [What is agentic automation?](https://agnt.gg/what-is/agentic-automation): Agentic automation puts reasoning inside your workflows. Classic automation executes rules you wrote in advance; agentic automation adds steps that can read, judge and decide when the rules run out.
- [What is an agent runtime?](https://agnt.gg/what-is/an-agent-runtime): An agent runtime is the environment where agents actually live: it executes their tool calls, manages memory and context, enforces permissions, and records what happened. Models think; runtimes act.
- [What is MCP?](https://agnt.gg/what-is/mcp): The Model Context Protocol (MCP) is an open standard for connecting AI models to tools and data. One MCP server — for a database, an API, a filesystem — becomes usable by any compliant client, ending the era of custom glue per model per tool.
- [What is local-first software?](https://agnt.gg/what-is/local-first-software): Local-first software keeps your data and computation on devices you control, treating the cloud as an optional accessory rather than the landlord. You own the files; sync and scale are choices.
- [What is a workflow engine?](https://agnt.gg/what-is/a-workflow-engine): A workflow engine executes multi-step processes reliably: triggers start runs, steps execute in order or in parallel, failures retry, and state survives restarts. It is the machinery under serious automation.
- [What is an AI operating system?](https://agnt.gg/what-is/an-ai-operating-system): An AI operating system is the layer that turns models into a working environment: agents as processes, tools as devices, memory as the filesystem, and receipts as the audit log. It is where AI work runs, not just where it is imagined.
- [What is agent verification?](https://agnt.gg/what-is/agent-verification): Agent verification is proving what an AI agent actually did: which sources it read, which tools it invoked, what it decided and what it changed. Without verification, delegation to agents is faith; with it, it is management.