PRODUCT GUIDE · OCTOBER 28, 2026

AgReady: A Self-Hosted AI Agent for Work You Control

An AI agent should help you get work done: understand a goal, plan the steps, use tools, work with files, remember useful context and produce something you can inspect.

AgReady brings these capabilities together in a self-hosted service. You can run the service on your own infrastructure, connect compatible models, provide authorized tools and manage work through user workspaces.

That combination is its central proposition: the agent capabilities you expect, with a deployment you control.

For people choosing an AI agent for research, documents and recurring business work, the question extends beyond whether it can answer a prompt. Where does the work run? Which model does it use? What can it access? How can you intervene?

AgReady makes these choices part of how you use the product.

The capabilities you expect from a working AI agent

AgReady covers the core workflow of an agent that acts on a goal rather than simply returning a conversational answer.

AgReady capabilities at a glance
What you need	What AgReady provides
Turn a goal into manageable work	Plans, task steps and plan revisions
Produce usable deliverables	Tools for files, documents, spreadsheets, presentations and PDFs
Work with connected services	HTTP tools and configured MCP connections
Carry useful context forward	Persistent memory that can be retrieved and managed
Reuse established ways of working	Skills containing instructions and supporting resources
Delegate parts of a larger task	Agent teams, subtasks and reports
Run recurring work	Persistent timers and configured external event triggers
Keep a person involved	Approval controls, interruption and continuation

These features belong together. A research task can involve gathering material, organizing findings, preparing a document and preserving useful context for the next task. A recurring workflow may also need a schedule, a connected data source and an approval boundary.

AgReady provides the building blocks for those workflows in one service. The specific tools and model still determine what a particular deployment can accomplish. Explore the core concepts.

Memory Skills and teams make work reusable

A useful agent should not require every task to begin from an empty page.

AgReady supports persistent memory for information worth carrying between sessions. Users can manage that information rather than treating every past conversation as an unquestionable instruction.

Skills serve a different purpose. They describe reusable ways of doing work, supported by instructions and resources. A team’s preferred report structure, research procedure or document workflow can become explicit guidance rather than something repeatedly explained in a prompt.

Agent teams provide another building block: dividing work into subtasks and bringing reports back together. This creates a way to organize larger tasks without claiming that adding agents automatically makes every job faster or more accurate. Read about memory and reusable Skills.

What the self hosted combination changes

Choose where the service runs

AgReady can run as a server-side service with user workspaces. The runtime does not have to live inside a single desktop session.

This matters when you want a stable place for files, task history, reusable instructions and recurring work. You operate the service and decide how it fits into your infrastructure.

Choose compatible models

AgReady supports connecting compatible model providers, including locally served models where the protocol and required capabilities are supported.

Model choice lets you match the model to the work and your deployment requirements. Text work, visual understanding and image generation may require different capabilities or configured model roles.

Using a local model does not automatically keep every workflow local. An external model, HTTP request or connected service can still transmit data. The relevant question is which components a particular workflow uses.

Keep recurring work on the server

Persistent timers support work that needs to recur, while configured external triggers can start work from authorized events.

For example, a recurring briefing could collect approved source material, prepare a summary and leave it for review. This is an illustrative workflow, not a reported test result.

Server-side persistence provides a place for that workflow to run. It still depends on the server, model and tools remaining available; approvals or other runtime conditions can delay execution. Review operational considerations.

Connect the tools your work actually uses

HTTP and MCP connections let an agent work with authorized services rather than relying only on uploaded text.

The important distinction is permission. An agent should receive the tools and access needed for a task, with appropriate limits on what it may do. A connected business system requires configuration and its own authorization; it does not become available merely because an agent knows its name. Explore tools and runtime assets.

Agent Ready means keeping people in control

The author’s vision of Agent Ready is a future in which agents operate business systems through APIs while people retain visibility and the ability to intervene.

There are two layers of control:

Inside AgReady: people set goals, inspect progress and use the available approval, stop and continuation controls.
Inside the connected system: people retain the system’s own interface and permissions to inspect records and intervene directly.

AgReady already provides tools for human oversight. The broader two-layer operating model is a design direction, not a claim that every external system is already integrated or controllable. Each connection needs to preserve the controls appropriate to that system. Read the Agent Ready vision and human oversight guidance.

How to read the comparison

Memory, Skills and multi-agent work are established product capabilities, not exclusive AgReady inventions. Deployment options also overlap: WorkBuddy documents an enterprise private-deployment offering.

AgReady’s case is the combination of familiar agent capabilities, a self-hosted service, user workspaces, compatible model choice and authorized tool access. It is not a claim that every alternative lacks each component.

Our comparison uses ITU-T F.748.46 as a reference framework for perception and cognition, planning, memory and execution. The standard helps organize the discussion; this article is a documentation-based product explanation, not an ITU certification or a common-condition performance benchmark. See the official recommendation.

AgReady, Codex, Claude Cowork and WorkBuddy: a capability comparison

Documentation checked as of October 28, 2026. These tables compare documented functions, not performance. No common-condition testing was performed. They are not a ranking, certification or claim of full compliance with ITU-T F.748.46.

The reference is the March 2025 edition of ITU-T F.748.46. Its perception and cognition, planning, memory and execution categories organize this comparison. The standard distinguishes required, recommended and optional items; not every item below is a mandatory requirement. Physical navigation and environmental sensing are outside this knowledge-work comparison.

Scope and evidence differ. AgReady is assessed from public and project-supplied product documentation, tool declarations and limited implementation checks, without independent testing of a matching deployment. Competitors are assessed mainly from official documentation, not live acceptance tests. Codex means the documented general knowledge-work entry points—not every ChatGPT feature. Claude means the Claude application and Cowork—not Claude Code or the separate Duet product. WorkBuddy means Tencent WorkBuddy.

Read each condition. Plans, versions, regions, administrators, models and connected services affect availability. The tables do not mean every capability is enabled in every default configuration. Supported means documented; Conditional requires a specified model, entry point, extension or permission; Partial covers only part of the item; Not supported is limited to an explicitly identified route; Unverified means insufficient evidence, not absence or failure to meet the standard.

Perception and cognition

Perception and cognition — documented capabilities
Capability	AgReady	Codex	Claude Cowork	Tencent WorkBuddy
Text semantics and multi-turn dialogue (§7.1.2–3)	Supported — depends on the configured model ^[2]	Supported ^[3]	Supported ^[11]	Supported ^[19]
Image input and understanding (§7.1.1–2)	Conditional — visual model required ^[2]	Conditional — available visual model required ^[4]	Supported — images and applicable PDFs ^[12]	Conditional — depends on model capabilities ^[21]
Text extraction from scanned documents (§7.1.1)	Conditional — OCR runtime dependencies ^[2]	Partial — reading text in images; full OCR workflow unverified ^[4]	Partial — visual PDFs; full OCR workflow unverified ^[12]	Unverified — complete scanned-document OCR workflow
Video input and understanding (§7.1.2)	Unverified — the general attachment route does not support it ^[2]	Unverified	Unverified	Unverified
Voice input (§7.1.1)	Unverified	Conditional — desktop entry point and availability ^[5]	Conditional — differs by entry point and mode ^[13]	Supported — built-in voice input ^[20]
Audio-file transcription (§7.1.1)	Unverified	Unverified	Unverified	Conditional — Local Whisper extension ^[24]
Text generation (§7.1.4)	Supported ^[2]	Supported ^[3]	Supported ^[11]	Supported ^[19]
Image generation (§7.1.4)	Conditional — configure an image model ^[2]	Conditional — available image tool ^[6]	Not supported for native photo/illustration generation; HTML/SVG graphics are separate ^[14]	Supported — built-in image tool ^[20]
Standalone audio-work generation (§7.1.4)	Unverified	Unverified	Unverified	Unverified
Knowledge reasoning (§7.1.5)	Supported — model and context dependent ^[2]	Supported ^[3]	Supported ^[11]	Supported ^[19]
Dedicated emotion-recognition system (§7.1.2)	Unverified	Unverified	Unverified	Unverified

A dedicated emotion-recognition system is an extra subdivision here; its unverified status does not rule out sentiment analysis in conversation. WorkBuddy documents video generation with sound, which does not establish video-input understanding or standalone TTS or music generation. Claude SVG or HTML visualizations are not native image generation.

Planning, memory and execution

Planning, memory and execution — documented capabilities
Capability	AgReady	Codex	Claude Cowork	Tencent WorkBuddy
Goal decomposition and multi-step execution (§7.2.1)	Supported ^[2]	Supported ^{[3, 7]}	Supported ^[11]	Supported ^[19]
Dynamic priorities and resource coordination (§7.2.2)	Partial — task and member assignments; global scheduling unverified ^[2]	Partial — subagent concurrency controls; global scheduling unverified ^[8]	Partial — parallel assignments; global scheduling unverified ^[11]	Partial — concurrency and duration limits; global scheduling unverified ^[23]
Plan adjustment after exceptions (§7.2.1–2)	Partial — plan revision and recovery; full automatic rescheduling unverified ^[2]	Partial — adjustment and continuation; full automatic rescheduling unverified ^[7]	Partial — user revision; full automatic rescheduling unverified ^[31]	Partial — connector reauthorization and retry; full automatic rescheduling unverified ^[26]
Within-session context (§7.3.1)	Supported ^[2]	Supported ^[3]	Supported ^[11]	Supported ^[19]
Cross-session memory (§7.3.2)	Supported — retrieval and management ^[2]	Conditional — separately stored local memory ^[9]	Conditional — cloud Cowork; local mode does not share that memory ^[15]	Supported — view, edit and turn off ^[22]
Tool and workflow execution (§7.4.1)	Conditional — HTTP, MCP and file tools ^[2]	Conditional — permissions and connected tools ^{[10, 30]}	Conditional — connectors and authorization ^[17]	Conditional — connectors and permissions ^{[25, 26]}
Human interaction and intervention (§7.4.1)	Conditional — approvals, stop and continue ^[2]	Conditional — permission mode ^[10]	Conditional — task and permissions ^[11]	Conditional — permission mode ^[25]

Timers are not dynamic scheduling, and subagents or concurrency limits do not establish optimal budget allocation, priority preemption or global resource optimization. AgReady documentation states that hard user and team budgets are not implemented. Approval and stop controls do not mean every write requires confirmation or that completed external actions can be undone. Permission-isolation strength, recovery reliability and failure compensation were not tested.

Additional usage and deployment conditions

These are additional product-selection conditions, not separate mandatory clauses of the standard.

Additional usage and deployment conditions — documented capabilities
Capability	AgReady	Codex	Claude Cowork	Tencent WorkBuddy
Multi-agent collaboration	Supported ^[2]	Conditional — applicable entry point and configuration ^[8]	Supported ^[11]	Supported ^[19]
Scheduled tasks	Conditional — user configuration and runtime conditions ^[2]	Conditional — local tasks require the machine online ^[16]	Conditional — cloud and local conditions differ ^[18]	Conditional — cloud and local conditions differ ^[23]
External event triggers	Conditional — configured and authenticated external event submission ^[2]	Unverified — general Codex entry point	Unverified — general Cowork entry point	Partial — project data-source webhooks; general task triggering unverified ^[20]
Discovering needs without a predefined goal	Unverified	Unverified	Unverified	Unverified
User-selected or local models	Conditional — protocol and model compatibility ^[2]	Conditional — specific client configuration ^[27]	Unverified — arbitrary third-party backends	Conditional — custom and local models ^[21]
Self-hosting the complete platform	Conditional — server deployment ^[2]	Unverified — a local client is not a self-hosted platform	Unverified	Conditional — enterprise private-deployment offering ^[28]
Web access and interaction	Partial — URL fetching; interactive browsing unverified ^[2]	Unverified — other ChatGPT entry points are not evidence here	Conditional — applicable connection and permissions ^[11]	Conditional — browser extension skill ^[29]

No performance conclusion: success rate, accuracy, completion time, stability and cost efficiency were not measured under a shared dataset, model, permission scope and budget. Use the tables to choose what to test, then record native capabilities, extensions and human assistance separately. An untested capability is neither a pass nor a failure.

Comparison sources

Numbered references link to the official documentation used for this comparison. Documentation can change; interpret each entry within the dated scope above.

Is AgReady the right fit

AgReady is worth evaluating when you want agent-assisted knowledge work and are willing to operate the service, configure models and tools, and define human approval boundaries.

Self-hosting brings responsibility for availability, updates and access management. Documented functionality also does not establish task success rates, security assurance or lower operating costs. Those require evidence from the deployment and work you actually intend to run.

Start with one small, verifiable task. Define the expected deliverable, connect only the necessary tools and inspect both the output and the agent’s actions. Add memory, reusable Skills or a schedule once that workflow meets your requirements.

Get started with AgReady and evaluate how much of your work you can delegate while keeping control of the environment in which it happens.

Optional Google Analytics