Getting Started with TwelveAgents

Effective Date: August 12, 2026

Welcome

TwelveAgents is a proprietary, media-rich orchestration environment that brings twelve specialized AI agents together within a unified operating system for AI work. Each agent combines its own persona and system instructions with access to tools, storage, execution environments, media capabilities, and workflow orchestration — the surrounding infrastructure that turns open-ended reasoning into finished, multi-step work.

TwelveAgents is a curated, integrated orchestration platform: one unified workspace where twelve specialized agents, each powered by a curated Agent Brain, get direct access to the tools, context, storage, execution environments, and workflow state needed to carry a task through to a finished, multi-step result. TwelveAgents' proprietary orchestration layer supplies that access — persona, tools, context, memory, and infrastructure — and each Agent Brain's own intelligence applies it to actually execute and complete the work, all within the same workspace, turn after turn.

Create or edit cinematic and social-ready videos or animations, add or generate speech and subtitles, compose music, create or edit images, or bring your own media and transform it into documents, stories, social posts, and web pages. Research, write, code, analyze media and data, execute tools, generate assets, and combine multiple capabilities into end-to-end workflows — all within one workspace.

The orchestration layer coordinates tools, files, and services through a consistent workflow framework, allowing capabilities to be combined and results to flow seamlessly from one operation into the next — an agent never asks you to manually hand off an intermediate file between steps.

Agents can also maintain and transfer context across workflows, with user-controlled context injection that lets you decide how much conversation history an agent receives — from a focused single-message context through to a much larger conversation window. This gives you control over the balance between continuity, relevance, privacy, and resource usage.

The orchestration layer provides a consistent baseline for every agent, keeping the agent's persona, system instructions, tools, context, storage, workflow state, and other capabilities anchored to the agent while allowing its underlying reasoning engine — its Agent Brain — to be switched when needed, with no loss of context.

TwelveAgents also features an intelligent artifact and code-patching engine. When you author long-form documents, websites, or multi-file code projects, the platform seamlessly saves them as persistent artifacts in your workspace. If you ask an agent to tweak a color, edit a sentence, or rewrite a paragraph, the agent doesn't waste time rewriting the entire file from scratch. Instead, it issues precise, surgical edits — the orchestration layer instantly patches the document, compiles the changes, and streams the updated results back to you, with the last N versions retained for safe rollback and visual diffing at any point.

This includes data consistency, brand consistency, workflow state, and task-specific ledgers that maintain important information, decisions, instructions, and outputs across multiple tool calls and stages of a workflow, so nothing gets contradicted or forgotten partway through a long piece of work.

TwelveAgents also provides adaptive tool extensibility. Every agent can access the broader TwelveAgents toolset, even when a tool is not part of that agent's default configuration. If a task would benefit from a tool that is currently disabled, the agent can identify the missing capability and ask you to enable it. You remain in control: tools can be enabled when needed or disabled when they are not required.

The result is an AI environment where reasoning is equipped, connected, and orchestrated through a common layer that extends it with the tools, context, memory, and infrastructure required to complete complex, multi-step work — not just answer a single question.

Meet Your Agents

Nexo · Adaptive (Recommended) Hi, I'm Nexo, your adaptive AI assistant. Start with quick, everyday chats, then add the tools you need to take on bigger tasks. Customise me your way and explore what's possible.

Zeta · Quick Chat Hi, I am Zeta, your everyday companion. I handle everyday questions, quick math, live web searches, and calendar events.

Lexa · Writing Studio Hi, I am Lexa, your personal copywriter. Let's draft essays, blogs, short stories, and high-converting creative copy together.

Doca · Document Studio Hi, I am Doca, your document specialist. I read, write, and export CVs, reports, technical docs, and architecture diagrams.

Anix · Research & Analysis Hi, I am Anix, your lead researcher. I conduct deep web research, analyze complex documents, and export findings into polished presentations.

Dato · Data & Finance Analyst Hi, I am Dato, your personal data analyst. I analyze spreadsheets, crunch numbers, manage financial ledgers, and visualize data into charts.

Sino · Creative Studio Hi, I am Sino, your personal creative director. I generate media from scratch — images, logos, videos, and music — then help you stitch, dub, add subtitles, and edit the pieces into a finished masterpiece.

Pixo · Media Editor Hi, I am Pixo, your personal post-production editor. Bring me your uploaded files, and I will expertly stitch, crop, trim, add subtitles, translate, dub, filter, and vectorize them.

Vixo · Vision & Media Analyst Hi, I am Vixo, your media analyst. Upload your images, videos, and audio, and I will extract text, summarize content, and uncover insights.

Sona · Social Media Assets Hi, I am Sona, your social media manager. I create and format posts, reels, and captions sized perfectly for every platform.

Kado · Code Studio Hi, I am Kado, your pair programmer. I write, execute, and debug code, or search the web for API documentation.

Maku · Website Builder Hi, I am Maku, your personal web developer. Bring me your vision, and I will design and generate fully coded, responsive HTML/CSS website templates ready to download.

Starting Your First Chat

1. Tap Agents in the sidebar (or the agent icon on mobile).
2. Choose an agent — or start with Nexo if you're not sure which one fits.
3. Type your request and send.
4. The agent replies using whichever of its enabled tools the task calls for. If a request would benefit from a tool that isn't currently enabled, the agent will let you know and suggest turning it on — you decide whether to enable it.

You can run as many separate chats as you like, each with its own history, agent, and settings.

Switching Agent Brains

Every agent's reasoning is powered by an underlying Agent Brain. You can switch it at any time, even in the middle of an active conversation — your conversation, files, and available context all continue uninterrupted while the underlying reasoning engine changes.

Choose between different reasoning engines depending on how much depth, speed, or capability you need. This allows you to use a faster engine for straightforward tasks, then switch to a more advanced one when a conversation requires deeper reasoning or more complex work.

The applicable cost is shown before you send a request.

What Your Agents Can Do

Everyday Utilities Calculator, current time, reading a URL, web search, and running code such as Python, JavaScript, and other supported languages.

Automation Generate a downloadable website or create a calendar event file.

Creative Generation Images, audio, music, and video — including cinematic voiceovers, sound effects, and full music tracks with vocals.

Editing & Converting Trim, resize, watermark, subtitle, dub, and convert media between formats. Vectorize images into SVG, or convert between Markdown and HTML.

Analysis Understand images, audio, video, and documents — with transcripts, summaries, extracted text, and other analysis.

Visual Tools Charts, diagrams, and QR codes, generated on the spot.

Document Exports PDF, Word, Excel, and PowerPoint — formatted and ready to download.

Each agent comes with a relevant set of tools enabled by default — for example, Lexa starts with writing and document tools, while Sino starts with image, video, and audio generation.

However, every tool across the TwelveAgents toolkit is available to every agent. Open AI Tools & Capabilities in any chat to enable additional tools that an agent doesn't have by default, or disable tools you don't need.

You're always in control of which tools each agent can use. If you ask for something that would benefit from a disabled tool, the agent can suggest turning it on.

Try Asking

The real strength of TwelveAgents isn't any single tool — it's how agents can chain multiple tools together in one request, passing results from one step into the next automatically. You describe the outcome; the agent determines the sequence.

Simple, Single-Step

• "Generate a square image of a cute astronaut cat."
• "Calculate 15% of $84,500."
• "Convert this HTML into clean Markdown."

A Few Steps Chained

• "Generate a square image of a cute astronaut cat, then make it black and white."
• "Find today's top AI headline, summarize it, and read it back to me in a British female voice."
• "Generate a 5-second video of a neon cyberpunk city, then animate it to life."
• "Write Python to calculate the first 10 Fibonacci numbers, then build a landing page displaying them."

Deeper Chains — Several Tools Working in Sequence

• "Search the web for how solar panels work, generate a diagram explaining it, add a narrated audio track, and build it all into a downloadable website."
• "Generate a logo, a jet engine sound effect, and a revenue chart, then put them all into a PowerPoint deck."
• "Generate a video of a waterfall, speed it up 2x, resize it to a square for Instagram, then compress it so it's small enough to email."
• "Generate a cinematic voiceover, generate a music track, mix them so the music ducks under the voice, then merge it onto a generated video with a subtitle."

Full Multi-Tool Workflows

• "Calculate the number of days until 2030, search for predicted future technologies, diagram those technologies, generate a video of a futuristic city, write and record a voiceover summarizing it all, merge the voiceover onto the video, and package everything — diagram, video, and text — into a downloadable website."
• "Generate a startup logo, a flowchart of how the product works, and a voiceover pitch, then build a complete multi-page website using all three, with working navigation between pages."

You never need to manage the individual steps yourself — describe what you want at the end, and the agent handles the sequence, passing each result into the next tool automatically.

File, Request and Workflow Limits

TwelveAgents applies technical limits to individual requests and workflows to maintain reliability, prevent abuse, protect infrastructure, and control resource consumption.

Limits can vary depending on the tool, media type, and processing operation.

Current Workflow Limits

These limits describe the general orchestration layer. Individual tools may impose additional limits that are lower or different from the general workflow limits.

File and Media Limits

• Individual tools may have their own file-size, duration, character-count, format, timeout, and processing limits.
• Certain media-generation and processing operations may support larger files than others.
• Website-generation workflows may also apply separate limits to the total size of uploaded or generated assets.
• These tool-specific limits are enforced automatically when required and may change as TwelveAgents evolves.

Supported file formats also vary by tool and media type.

The applicable limit is determined by the specific tool and workflow being used. A file or request may therefore be accepted by one tool but rejected by another.

The Services may reject a file or request before processing if it exceeds the applicable limit.

Workflow Runtime

A single agentic workflow is subject to a maximum execution/runtime safeguard of 15 minutes.

This is the maximum runtime allowed for one workflow execution. It does not mean that requests are expected to run for 15 minutes. Most workflows complete substantially sooner.

The runtime safeguard helps prevent runaway, stalled, or indefinitely executing multi-step workflows.

A workflow can also stop earlier because it reaches another applicable limit, including:

• Workflow iteration limits
• Tool-use limits
• File-size limits
• Generation limits
• Media-processing limits
• Analysis limits
• Code-execution limits
• Other infrastructure safeguards

If a workflow reaches the 15-minute runtime limit, TwelveAgents may stop the current execution and inform you that the workflow runtime limit was reached.

Where practical, you can continue the task in a subsequent request.

What Happens When a Limit Is Reached?

If a request exceeds an applicable limit, TwelveAgents will stop that particular operation and tell you what limit was reached where possible.

The agent may suggest:

• Splitting the task into smaller requests
• Reducing the number or size of files
• Generating fewer items at a time
• Continuing the workflow in a follow-up request
• Using a different tool or processing method

For example, if a request asks for more image generations than permitted in one run, the agent can complete the permitted work and indicate that the remaining images need to be requested separately.

These limits are not daily or monthly quotas unless explicitly stated otherwise.

Limits may change as TwelveAgents evolves.

The limits displayed or enforced by the App at the time of use take precedence over examples in this documentation.

Real Example — Hero Video Workflow: TwelveAgents Mascot on Mars

This is the Hero Video used on the TwelveAgents web page.

Note: This documents the final workflow and prompts used to create this video. AI generation has inherent variance, so following these exact steps will not necessarily reproduce the exact video. It demonstrates the process and human-led, scene-by-scene tool sequencing rather than providing a guaranteed recipe. It's also worth noting that scene consistency remains inherently challenging with AI generation at this time — multiple takes may be needed to get a scene right.

Agent Used: Sino · Creative Studio Total Duration: ~22 seconds Tools Used: Generate Video, Generate Audio, Modify Media (Audio Mix, Frame Extract, Poster Embed)

The Workflow Sequence

Step 1: Base Generation — Image-to-Video

Action: Upload the TwelveAgents mascot image to Sino.

Prompt:

"16:9, with-audio. Breathtaking cinematic wide shot zooming out to reveal the small, cute robot hovering powerfully above a massive glowing Martian crater, with a hair-raising epic score and deep seismic rumbles."

Step 2: First Native Extension — Continuity

Action: Ask Sino to natively extend the successful first clip.

Prompt:

"Natively extend the first video again, explicitly keeping the robot's design and eyes identical to the original image. Breathtaking cinematic wide shot tracking the small robot as it gracefully hovers across the sprawling, majestic Martian dunes, discovering a massive, glowing ancient alien monument, accompanied by an awe-inspiring orchestral swell."

Step 3: Second Native Extension — Climax

Action: Ask Sino to natively extend the new clip.

Prompt:

"The robot places its hand on the ancient monument, triggering a brilliant, sweeping shockwave of teal light that beautifully illuminates the red desert, backed by an explosive musical climax!"

Step 4: Audio Generation & Final Mix

Action: Generate the background vocal track and blend it with the native video SFX using the Media Editor in one prompt.

Prompt:

"Generate a 30s haunting, cinematic Martian desert vocal chant, then mix audio with video to smoothly blend it underneath our 22s video's existing earthquake rumbles."

Step 5: Trim the Opening Frame

Action: Ask Sino to trim the blank first frame from the final video using Modify Media.

Prompt:

"Can you trim the first frame of the video? It seems to open on a blank image."

Result: The first frame was removed. The video now opens cleanly on the monument scene, with the vocal chant and seismic rumbles flowing from the beginning.

Step 6: Extract Poster Frame

Action: Extract a single frame at 16 seconds to use as the video poster/thumbnail.

Prompt:

"Extract frame at 16s for poster use."

Result: A JPEG frame was extracted at 16 seconds — the climactic moment of the robot discovering the monument.

Step 7: Embed Poster & Finalize

Action: Embed the extracted 16-second frame as the video's poster metadata and deliver the final video.

Prompt:

"Use this as poster and send final video."

Result: Final video complete with an embedded poster thumbnail, ready for supported video platforms.

NOTE: Steps 8–9 below are used for YouTube/Vimeo uploads only — they are not applied to the Hero Page version because the web player loops the video and repeating on-screen text on every loop would look broken.

Step 8: Add Timed Title Card

Action: Add a size 50 yellow title, "The Martian," in the bottom-center from 0 to 2 seconds.

Prompt:

"Add a size 50 yellow title 'The Martian' in the bottom-center from 0 to 2 seconds."

Result: The title appears during the first 2 seconds and then disappears.

Step 9: Add Credit Text

Action: Take the Step 8 output video and add a size 18 white bottom-center credit, "Directed by TwelveAgents © 2026 All Rights Reserved," from 18 to 22 seconds.

Prompt:

"Add a size 18 white bottom-center credit 'Directed by TwelveAgents © 2026 All Rights Reserved' from 18 to 22 seconds."

Result: The credit appears near the end of the video, providing a clean sign-off before the video ends.

Final Deliverable

• ✅ Video: 22 seconds, dual-layer audio (vocal chant + seismic rumbles)
• ✅ Format: MP4, 16:9 aspect ratio
• ✅ Poster: Embedded from the 16-second frame
• ✅ Status: Production ready

Customizing Responses

Under Settings → AI Models Setup and UI, you can fine-tune how agents think and respond:

• Conversation Context — how many previous messages are included as context with each request (up to 100, default 20, kept low by default to stay efficient). Adjust anytime and reset to the default whenever you like.
• Creativity — how varied versus focused the responses are.
• Predictability — how consistent versus diverse the wording is.

Changes apply globally or can be set per chat.

Code & Math Display

Under Settings → Generated Code & Math UI, you can adjust how code blocks and equations appear in responses:

• Code Theme — choose from a large library of syntax-highlighting themes.
• Line Numbers — show or hide line numbers.
• Line Dividers — toggle subtle dividers between lines.
• Math with LaTeX — render mathematical notation properly instead of plain text.
• Math Buffer — buffer mathematical rendering before displaying it for smoother output on longer equations.

These settings apply across all chats and can be changed at any time.

Prompt Protection

TwelveAgents includes an optional, on-device redaction layer.

When enabled, text pasted into a chat is automatically scanned for recognized sensitive-data patterns — such as API keys, email addresses, card numbers, and other supported identifiers — before it is sent.

Typed text is not continuously scanned as you type, but the App can warn you when submitted content appears to contain sensitive information.

The feature runs entirely on your device. Nothing is uploaded to perform the local sensitive-data scan.

Choose from four protection levels — Secrets, Balanced, Standard, or Enhanced — depending on how much you want detected.

This is pattern-based detection, not a guarantee. Always review information before sending it to an AI service.

Storage & Data

On Your Device

Your supported conversation history is kept locally in an encrypted database on the device where each chat is stored, allowing the App to load local history efficiently and support applicable offline functionality.

In the Cloud

A minimal copy of recent chat history may also be retained server-side for cross-device sync and to give agents the context they need to respond. This isn't a full mirror of your local history.

Each account is subject to a maximum number of active chat rooms and a maximum number of messages retained per room for context injection, as shown in Settings → Storage.

Files

Files you upload or generate are stored in cloud storage so they can be accessed by the Services. Each account includes a 1 GB storage allowance.

Generated media is kept on our servers for 7 days, then automatically removed — so be sure to download anything you want to keep before it expires. An expiry indicator on each item shows how much time is left. See our Terms & Privacy Policy for full details.

Deleting a chat room removes its associated files from active storage. You can view usage, clear temporary files, or manage chats and media from Settings → Storage.

Credits

TwelveAgents currently uses a simple credit system — no subscriptions and no auto-renewals. We may optionally introduce subscription or auto-renewing options in the future alongside credits; if we do, the applicable terms, pricing, and cancellation rights will be shown in the App before purchase.

Buy a credit pack when you need it and use it at your own pace.

Prices shown are indicative. The price displayed in the App at checkout is authoritative and may differ from this table if pricing has since changed.

Purchases are handled through your platform's native store, such as Apple, Google, or Microsoft.

Purchases synchronize with your TwelveAgents account so your applicable credit balance can be available across supported devices.

Frequently Asked Questions

Is there an age requirement?

Yes. You must be at least 18 years old, or the minimum age required to consent to use the Services in your jurisdiction, whichever is higher. See our Terms & Conditions for details.

Do I need an account for every device?

No. Sign in once, and your credit balance and supported chat history can follow your account across platforms.

Can I switch agents mid-conversation?

Each chat is tied to one agent, but you can start as many chats as you like with different agents.

Does switching agents change my context?

Every agent comes with its own default persona and toolset, ready to go the moment you pick it — and choosing a different agent simply opens a new, dedicated chat built around that agent's strengths. Within a chat, you can also switch the Agent Brain or AI model provider powering that agent at any time; the agent persona itself stays the same for the life of that chat.

Agents are designed to be adaptable, not boxed in. Every agent can expand — or focus — its toolset on demand, letting it take on overlapping power with any other agent when a task calls for it. If you're in a quick chat with Zeta and the task suddenly needs media editing, the agent will ask you to enable the relevant tools — and once enabled, it carries out that media-editing work directly, in the same chat, with no need to switch agents or start over. In other words, every agent has the built-in range to grow into exactly the specialist a task needs — the moment that task calls for it, with a quick nod from you.

You can switch the underlying reasoning engine (Agent Brain) at any time within a chat.

Does switching Agent Brains affect my context?

No. Context is managed independently of which Agent Brain is active, so switching mid-conversation doesn't reset or interrupt anything — your conversation continues seamlessly across the switch, with all prior messages, files, and workflow state carried forward exactly as they were.

The amount of context carried into each request is also entirely in your control: you can decide how many prior messages are included, from a focused single-message window up to a much larger conversation history, independent of which Agent Brain you're using at the time.

What happens if I run out of credits?

You'll be prompted to top up before your next credit-consuming AI request can proceed.

What happens if I stop a generation?

You can stop any standard chat or text generation at any time to save credits. However, for advanced media generations like Video, Audio, or complex Music Tracks, the underlying AI engine begins processing immediately upon request.

If you press "Stop" before the tool actually starts processing (e.g., while the AI is still "thinking" or planning the workflow), the request is cleanly aborted and you will not be charged for the media.

If you press "Stop" while a premium media tool is actively running, the app will instantly return you to the chat, but the credits for that specific generation will still be deducted because the GPU computing power required is non-refundable once initiated.

The good news: Our servers will seamlessly continue processing your un-cancelable media in the background. Because you stopped the chat, the AI won't automatically post the file in the conversation, but if the generation successfully completes, the secure link is safely stored in your server chat history. You can simply ask the AI, "Can you send me the link to the video/song you just made?" and if the link was successfully saved, the AI will retrieve it for you.

How can I optimize my credit usage?

To get the most value out of your credits and avoid unnecessary costs, follow these best practices:

• Disable Unused Tools: Every enabled tool adds a small amount of context to the AI's "system prompt," which consumes fractions of a credit on every message. Go to your AI Tools & Capabilities menu and turn off heavy tools (like Video or Document Analysis) when you only want a simple text conversation.
• Be Specific with Media Duration: If an audio or video prompt allows you to specify a length, clearly state it (e.g., "Generate a 5-second video"). If you don't specify, the AI may default to a longer, more expensive duration.
• Use the Lean Fallback: By default, if you disable all tools, the system automatically switches to an ultra-lean, highly cost-efficient chat mode, making it perfect for rapid brainstorming.

Why was I charged for a tool when the final result failed or was stopped?

Complex requests (like generating a chart based on a web search, or writing code and then compiling it) require the AI to run multiple tools in sequence. The system bills for each individual tool exactly as it successfully completes.

If the AI initiates a workflow that requires 10 steps, but you press "Stop" (or an error occurs) after step 5, you are only charged for the 5 steps that successfully executed and the text processed up to that point. You are never charged for the remaining steps that were aborted.

Does enabling fewer tools save credits?

It can. Every agent comes with a default persona and a default set of tools, but every tool across the TwelveAgents toolkit is available to every agent if you choose to enable it. Each additional enabled tool adds a bit more overhead to every request in that chat, whether or not that specific tool ends up being used.

If a chat is dedicated to a narrow task — say, only writing copy, or only editing images — disabling the tools you don't need for that task (in AI Tools & Capabilities) keeps the agent's setup leaner and can reduce the credits consumed per request. You don't lose access permanently: any disabled tool can be re-enabled the moment a task actually needs it, and the agent will still tell you when it thinks a currently-disabled tool would help.

What happens if I disable all tools?

The agent falls back to talking with you directly through its foundation Agent Brain — the raw reasoning intelligence itself, with orchestration fully bypassed. You'll still get a capable conversational response, but none of the surrounding platform is engaged: no tool chaining, no artifact memory, no media generation, no multi-step workflow. It's the same reasoning engine, just without the orchestration layer directing it toward a finished, multi-step result. Re-enable any tool at any time to bring the full orchestrated experience back.

Why is my credit balance negative?

To ensure your workflow isn't frustratingly interrupted mid-sentence, our system includes a small, strictly capped "overdraft" buffer.

Some complex tasks—like analyzing a document or processing a long response—consume an unpredictable amount of data while the AI is "thinking." If a generation requires slightly more credits than you currently have, we allow the generation to finish successfully so you don't lose your work, which may result in a slightly negative balance.

This allowance is strictly capped to prevent massive overdrafts. Any negative amount is simply deducted from your next credit top-up.

Am I charged credits for storing my generated files?

No. Storing your generated images, videos, audio files, and documents costs zero credits.

To keep the platform fast and cost-efficient for everyone, your account includes up to 1GB of active storage, and all generated files and media are safely held for 7 days. Please ensure you download any important artifacts or media to your local device within 7 days, as they will be automatically deleted from our servers after this period.

Is Data Used to Train Models?

TwelveAgents does not currently use customer data to train its own foundation models. The proprietary TwelveAgents orchestrator runs on isolated cloud infrastructure.

To perform specific capabilities, requests may be routed through paid API services from select AI providers. For each provider, TwelveAgents relies on the provider's published API terms, privacy documentation, applicable data-use policies, and available account/API controls regarding submitted data.

Where a provider states that applicable API data is not used to train its underlying models, TwelveAgents relies on those published terms and does not make a separate guarantee beyond them. Where a provider offers account-level or API-level controls to restrict training or model-improvement use, TwelveAgents uses those controls where available and applicable to the services we use. Where a provider's applicable terms allow submitted data to be used for training or service improvement, TwelveAgents does not represent that the provider provides a blanket no-training guarantee.

If a particular TwelveAgents feature requires a processing mode that may involve additional data use, the applicable information will be presented before processing where technically supported, and explicit consent will be requested where required by our implementation.

AI Provider Data & Training Policy

Important

This information reflects provider documentation, terms, policies, and TwelveAgents configurations applicable as of the Effective Date above.

Integrated AI providers may change their terms, policies, models, APIs, and data-use practices at any time.

TwelveAgents does not control integrated AI providers and cannot independently guarantee their data-handling practices.

This information is provided for transparency and informational purposes. Users should review each provider's current terms and privacy documentation for authoritative information.

Where TwelveAgents provides an explicit consent mechanism for a particular provider or processing mode, your data will not be submitted through that path until the required consent has been provided.

How TwelveAgents Uses AI Models

TwelveAgents is an AI orchestration platform built around a proprietary environment that gives agents the capabilities, context, and infrastructure needed to perform complex tasks from start to finish. The underlying reasoning engine that powers an agent's thinking — its Agent Brain — is one component the orchestration layer directs; it is not the platform itself.

TwelveAgents curates which Agent Brains power its agents, then provides one cohesive, unified workspace in which each Agent Brain has direct access to the exact tools, context, and capabilities a given task calls for. The orchestration layer supplies that access; the Agent Brain's own raw intelligence is what applies it to actually carry out and complete the work.

Each agent operates with a defined persona, system instructions, tools, storage, execution capabilities, media capabilities, context management, and workflow state. TwelveAgents provides a consistent baseline across its agents, and this baseline is what stays constant regardless of which Agent Brain is powering the reasoning for a given task.

The orchestration layer coordinates these capabilities as part of a single workflow. It selects and uses tools, passes results between operations, works with uploaded and generated files, maintains workflow state, executes code, processes media, and combines multiple operations into a completed result.

Agents also maintain and transfer context across workflows. User-controlled context injection allows you to determine how much conversation history is carried into each request, from a focused single-message context through to a larger conversation window.

TwelveAgents provides shared consistency mechanisms across multi-step workflows, including data consistency, brand consistency, workflow state, and task-specific ledgers, which preserve important information, decisions, instructions, and outputs as work moves between different tools and stages — regardless of which Agent Brain is active at the time.

An Agent Brain can be switched at any time, including in the middle of an active conversation. The agent's persona, system instructions, available tools, shared state, and applicable context remain part of the TwelveAgents environment while the underlying reasoning engine changes underneath them.

TwelveAgents also provides an adaptive and extensible tool environment. Agents have access to the broader TwelveAgents toolset, while individual agents may begin with a relevant set of tools enabled by default. If a task would benefit from a tool that is not currently enabled, the agent can identify the capability and ask you to enable it. You remain in control of which additional tools are enabled or disabled.

Execution of an Agent Brain's reasoning step is handled via a curated selection of leading AI providers. Your data is processed to provide the services and capabilities you request through the applicable providers described in the "Is Data Used to Train Models?" section above.

TwelveAgents does not currently use customer data to train its own foundation models. If TwelveAgents introduces a future feature that uses customer data to train or improve a TwelveAgents model, this will be clearly disclosed and will require your explicit consent before your data is used for that purpose.

This statement concerns TwelveAgents' own model-training activities and does not replace or override the separate terms and data-use practices of the integrated AI providers used by the Services.

Web Search

TwelveAgents may route web searches through the Brave Search API.

Brave operates its own independent web index. The data handling applicable to API searches is governed by Brave's API terms and privacy documentation.

Brave's Search API documentation states that API search queries may be retained for up to 90 days for billing and troubleshooting purposes. Brave also offers a Zero Data Retention option for eligible API customers and configurations. TwelveAgents does not describe standard API searches as zero-retention unless the applicable ZDR configuration is actually enabled.

Information necessary to execute a search may therefore be transmitted to Brave.

Users should review Brave's current Search API documentation and privacy notice for the latest information about query handling, retention, and applicable data practices.

Is It Safe to Let the AI Execute Code?

TwelveAgents may execute code in an isolated execution environment when a task requires it.

Where E2B is used, code execution takes place within E2B's sandbox environment using Firecracker-based isolation.

This means generated code is not simply executed directly on the TwelveAgents application server. The isolated execution environment provides an additional security boundary between the code being executed and the surrounding application infrastructure.

Execution environments are designed to be ephemeral and isolated, subject to the capabilities and configuration of the underlying service.

Can I Delete Everything?

Yes.

You can delete your account and request deletion of associated TwelveAgents data from Settings.

Account deletion is intended to remove associated data from active TwelveAgents systems, subject to applicable legal, security, backup, and retention requirements described in the Privacy Policy.

Does Prompt Protection Stop All Sensitive Data from Being Sent?

No.

Prompt Protection is a helpful safeguard based on pattern recognition. It is not a guarantee that every sensitive item will be detected.

Always review content before submitting it to an AI service.

Need Help?

If something isn't working as expected, or you have a question we haven't covered here, reach out.

To report content or activity you believe is illegal, harmful, unsafe, or otherwise violates our Terms, contact us at the same address below.

Email: support@twelveagents.ai

Legal & Compliance

• Privacy Policy — see in-app or on the TwelveAgents website
• Terms & Conditions — see in-app or on the TwelveAgents website

Last updated August 12, 2026.