Questions, answered.

Yes, you can try Prismr for free for 2 weeks.

Prismr runs on Mac, Windows, and iOS (iPhone and iPad). Linux and Android are coming soon. The experience is adapted per device so it feels usable, not squeezed.

1. Install Prismr.
2. Start a project, use Quick Chat, or use Prismr the same way you would use any chat or LLM tool.
3. Connect your AI provider API key in Settings, or add a small balance to your Prismr.ai LLM account.
4. Add files, attach a PDF, or start chatting.
5. Save important context as Memories, history or artifacts.

$49 per user per year during the introductory special. That's it.

Prismr supports BYOK (bring your own key) to support your favorite LLMs directly, or you can pay for usage through Prismr. There are no monthly fees; you just pay for what you use, and you'll be surprised how low that is for most users.

No, it is not required. You can bring your own API keys if you prefer. But if you'd rather not manage API keys, you can simply pay for usage through your Prismr account instead.

Your trial lasts 2 weeks. After that, Prismr is $49 per user per year during the introductory special.

No. We deliberately don't offer a permanent free tier. We'd rather win on product quality than on giveaways. The subscription is our floor, and it keeps us accountable to paying users rather than optimizing for conversion funnels.

Prismr supports Claude (Anthropic), GPT (OpenAI), Gemini (Google), OpenRouter (350+ models), and LM Studio (local models on desktop).

Yes. You can switch from Claude to GPT to Gemini in one click, or simply @mention the model by name inside the chat (for example, typing @opus47 immediately wakes up Claude Opus 4.7 for that turn). The conversation keeps going. This lets you use the best model for each task, and even lets them critique each other until you get the best results.

OpenRouter gives you access to over 350 models from dozens of providers through a single API key. Prismr hosts models both through OpenRouter and natively with the LLM companies directly, picking whichever path delivers the best price-to-performance ratio for you.

Yes. On desktop, Prismr supports LM Studio for local models. This enables completely private AI interactions without sending data to external servers.

You pick an intent, not a model. Prismr offers three tiers: Agentic ("give me the deepest, most capable mind for this"), Smart ("do this well without burning money," the default), and Fast ("good enough, quickly"). Prismr also ships presets for AI team combos so you can spin up a pre-built multi-model lineup (for example, a researcher plus a critic plus a writer) with one click. You can always override and pick a specific model yourself.

No, and that's why we benchmark. Our internal evaluations have repeatedly shown that price is a poor proxy for quality. A cheaper model often ties or beats a flagship on specific task types, and newer doesn't always mean better. The tiers and presets are there to guide your choice, but the final pick is always yours.

Yes. Prismr supports file attachments and automatic PDF text extraction for use in chat. It shows extraction stats so you can verify what was captured.

Markdown, PDF (with text extraction), Word documents (.docx), Excel spreadsheets (.xlsx), CSV files, and images. More coming.

When you attach a PDF, Prismr automatically extracts the text content. It shows you stats like original file size versus extracted text length. This transparency helps you spot when a PDF did not extract well.

Yes. PDFs and images can be viewed directly within Prismr without opening external applications.

Project Memory is where you store key facts, preferences, and context that should carry forward across chats inside a project. It reduces repeated explanations and keeps work consistent over time.

Chat history is transient, unless you save it or create artifacts from its contents. It resets when you close a conversation. Memories are permanent. They persist across sessions and get automatically included in future conversations.

Yes. You have full control. You can add, edit, tag, archive, and remove memories so your project knowledge base stays accurate.

In a file in your project folder. This means memories sync with your project files across devices.

No, not by default. Each project is sealed. Its context, decisions, and agent notes stay inside that project's folder. There are three deliberate ways to cross-pollinate:

1. User Memory is personal-to-you and applies everywhere automatically (your preferences, communication style).
2. You can explicitly tell the agent to read another project as reference material for a one-off task.
3. You can promote a reusable artifact to Team Memory so it's available across all projects.

Prismr never silently copies information between projects.

No. Prismr uses your filesystem as the memory layer. Memories are plain markdown files you can open, read, and edit yourself. No black box, no vendor lock-in. If you want to inspect exactly what the agent "knows," you can open the file.

Expert Mode analyzes your request and helps assemble better context and prompts. It reduces back-and-forth and helps you reach better outputs faster. Think of it as a built-in prompt engineer.

It analyzes your intent, scores your files for relevance, suggests relevant memories and documents, and rewrites your prompt to get the best possible answer.

Yes. You can preview how Expert Mode enhanced your prompt and what context will be included before you send.

Your choice. Prismr works with iCloud, Dropbox, Google Drive, OneDrive, or your local filesystem. We never store your documents on our servers.

Yes. On desktop, Prismr supports local filesystem projects. This is ideal for privacy-sensitive work, offline work, or teams with their own sync systems.

Yes. Projects sync through your cloud storage provider. Start on your Mac, continue on your iPad. Same workspace, any device.

Yes. You can move projects between storage providers. Your files are portable.

Yes. Your files live in your storage. AI requests go directly to providers (Anthropic, OpenAI, Google, etc.). Prismr (the company) does not see your content.

When you use paid API models, your data is generally not used for training. Free models from some providers may use your data for training purposes. Check each provider's terms for details. You control where your data lives.

Partially. With local storage and LM Studio models (desktop), you can work without internet. File operations work offline with local storage.

No. Secrets never enter the model's context window. Prismr uses opaque references: the AI sees a placeholder, and the real secret is substituted in at execution time, just before the action runs. This means your credentials are never seen or logged by Anthropic, OpenAI, Google, or any other model provider.

Whichever AI provider you pick. When you choose Claude, your request goes to Anthropic. When you choose GPT, it goes to OpenAI. The model picker is itself the disclosure. Prismr doesn't route your data to third parties you haven't chosen.

Yes. Through file sharing (for example, Dropbox, Google Drive, or any file-share system of your choice), you can share entire projects, manage collaborators, and control access levels.

You can add comments on documents, tied to specific text selections. Comments include author initials and timestamps. This supports review workflows inside Prismr.

Yes. Comments track the author using their license email and display initials. Timestamps show when each comment was created.

Yes. Using whichever file-sharing service you've connected (Dropbox, Google Drive, OneDrive, iCloud, or similar), you can invite collaborators and set access levels (owner, editor, or viewer).

Not today. SSO and SOC 2 are on the roadmap for a future enterprise tier, not in the current product. If you need those today, talk to us about the enterprise roadmap.

Quick Chat lets you start chatting immediately without creating a project. It is ideal for quick questions or brainstorming. When a conversation becomes valuable, you can save it into a project.

Yes. AI responses can be saved as formatted artifact files within your project structure. One click turns a chat response into a permanent, editable document.

Yes. Prismr estimates token usage and shows context window consumption before you send. No surprises.

Yes. You can import chat history from popular LLMs. Start a conversation in ChatGPT, Claude, or Gemini, continue it in Prismr.

You can save frequently used prompts as reusable templates. Write once, use many. This saves time and ensures consistency.

ChatGPT, Gemini, and Claude are one-on-one chats. You talk to one AI at a time, alone. If you want a second opinion from a different model, you leave that app, open another one, and re-explain yourself. The context of your original conversation is gone. Prismr was built for real work. Here's what's different:

The wisdom of the full team of models

Claude, GPT, Gemini, and others sit in the same thread. They hear each other, challenge each other, and build toward a conclusion you can actually act on.

Your whole team shows up

Prismr is a shared workspace, not copies of conversations pasted around Slack. Colleagues and AIs participate in the same group chat.

Prismr gets real work done

Agents don't just answer. They execute missions, draft documents, control a browser, and hand you deliverables for approval. Chat apps reply. Prismr finishes.

On top of that, project memory persists across sessions so context never fades, your files are grounded in real content (not summaries), and the workspace syncs across desktop and mobile so the work follows you.

Multiplayer by design

Prismr is the only platform in this group built for teams from day one. Cowork, Codex, and Perplexity Computer are fundamentally single-user products. Prismr lets multiple humans and multiple LLMs work in the same live chat, with real-time group conversations, direct messages, voice and video, and shared agent sessions that the whole team can see unfold.

Your choice of AI model

Prismr curates a selection of models so you always have the right tool for the job. That includes the major names you already trust (Claude, OpenAI, Gemini) alongside powerful models you may not have discovered yet. Because Prismr helps you match the model to the task, you get better results more cost-effectively than staying locked to a single AI family. A lightweight, fast model handles quick tasks well; a more capable model is the right call for high-stakes outputs like board presentations or legal summaries.

An Agent Office that adapts to the job

Prismr intelligently selects the right execution model for each task. Some jobs are best handled sequentially, with one focused agent working step by step. Others benefit from parallel execution, with multiple agents tackling different workstreams at the same time. Prismr supports both approaches and chooses between them based on what the work actually requires, so you get the right structure rather than a one-size-fits-all loop.

Relentless fact checking

Every Agent Office output goes through multiple independent review cycles before it lands. The reviewer has one job: find what is wrong, incomplete, or unsupported, and either trigger a rescue pass or sign off with a hard COMPLETE. You do not get a confident-sounding answer; you get a verified one.

A browser agent with its own identity

The Prismr browser agent operates in a fully sandboxed, isolated browser session with its own dedicated persona and email address. It works independently from your real accounts, keeping your personal identity separate from agent activity. This means safer execution, cleaner accountability, and a clear record of what the agent did and why.

Desktop and iOS, today

Prismr ships a fully working iOS and iPad app alongside its Mac and Windows desktop apps. Most competing agentic systems lack mobile functionality or offer insufficient mobile functionality; Prismr covers all three.

Transparent by design

Prismr preserves every artifact your agents produce, so outputs are available for future reference, review, and auditing. And rather than hiding agent activity in a black box, Prismr gives you a visual view of your Agent Office: a live picture of which agent employees are working, on what, and for whom. You always know what is happening and who is responsible.

Open System / No Lock-in

Your files stay on your own storage, everything belongs to you, and your memories and artifacts are portable to other agentic platforms. We never hold your data hostage by hiding it from your local tools.

Prismr is an action platform, not a research tool. The agent completes real work on your behalf, including research missions, form submissions, reservations, orders, document drafting, and data gathering, and shows its work through an approval flow. The core product surface is the proposal card: the agent says "here's what I'm about to do," and you say "go ahead."

Business users who need leverage without hiring more staff. That includes solo operators, small teams, and larger organizations alike, typically Heads of Ops, Marketing, Finance, and similar roles who are stretched thin on technical resources. Prismr is designed for people who work on outcomes, not code, regardless of company size.

Both, depending on the job. Simple tasks run step-by-step with a single agent. Complex missions fan out to multiple agents working in parallel, each on a different piece. You can watch them work in real time inside the desktop app, or hide the detail and trust the process.

Yes. Visibility is core to how Prismr works. Three surfaces answer three different questions:

• What's happening now? Mission Control inside the desktop app (always on, green/yellow/red status at a glance).
• What needs me? Notifications (deliberately quiet, only firing for decisions, completion, blockers, or budget thresholds).
• What did we conclude? The "Brief me" button on any mission, which produces a one-page summary in your preferred style.

It depends on the task. Before a mission starts, Prismr proposes an estimated time, cost, and confidence level. You can accept the default or adjust (go deeper, keep it quicker). Mid-flight, if a short mission clearly needs more depth, it can upgrade without starting over.

Two layers. First, the system asks itself "did I complete the plan?" An internal reviewer either marks the mission COMPLETE or flags follow-up tasks, with a hard cycle cap so it can't loop forever. Second, you look at the output and decide whether it's deep enough. If not, one click kicks off a follow-up run that builds on what's already been done. The machine can verify execution; only you can verify whether the depth matches your intent.

No. Anything with real-world consequences (sending an email, submitting a form, making a payment, writing a file outside the agent's sandbox) requires an explicit approval from you first. The agent can read freely to gather context, but it can only write into a sandbox directory until you promote the changes. The approval card is the consent boundary of the entire product.

AI agents are powerful but imperfect. Models can misread instructions, misinterpret a web page, or be nudged off course by content they encounter while browsing. Our job is to keep you in the loop: the approval card shows you what the agent is about to do before it does it, we maintain an audit trail of every action taken, and we design defaults to be cautious. Your job is to treat agent output the way you'd treat work from a capable but new team member, review before you approve, especially for anything with financial, legal, or reputational consequences. Exercise the same care you would with any other tool that can act on the world.

Prismr tries to warn you when a site may prohibit automation, but doesn't maintain a hard refuse list, because a refuse list would imply we've vetted every site not on it. Detection isn't always possible, so you should also use your own judgment. You decide whether to proceed, with any warnings we surface visible on the approval card.

Yes. Every mission generates a structured audit trail (tool calls, URLs visited, actions taken, approvals granted, timestamps) with secrets automatically redacted. There are three levels of detail depending on your tier: a readable replay for consumers, full timestamps and cost data for business users, and tamper-evident, exportable logs for enterprise/regulated users.

This is the hardest open problem in the agent space, and we're honest about it: nobody can claim to be injection-proof. Our approach is to assume injection will eventually succeed and engineer so the blast radius is bounded. The load-bearing defense is the approval gate: even if the agent is tricked into wanting to do something dangerous, it still has to ask you before it acts. We also maintain a library of adversarial test cases that re-run on every code change.

Yes. We publish our threat model publicly. We'd rather invite researchers to poke holes in a defensible posture than hide behind a marketing claim of "unhackable."

No, not without your consent. By default, code executed by the agent can only reach websites you've explicitly allowed. When it needs a new destination, it asks: "I need to call api.example.com, allow for this project?" Power users can flip to an "allow-by-default with blocklist" mode. Either way, every outbound call is logged through an audit proxy. No exceptions.

MCPs (Model Context Protocol plug-ins) connect the agent to external tools and data sources, for example a UK property database, a CRM, or a specialized research source. They extend what Prismr can do.

Prismr uses a tiered trust model based on where an MCP comes from:

• First-party (built by Prismr): trusted.
• Verified (identity-verified publisher, audited): trusted with audit logging.
• Community (unverified): treated as untrusted data. Output is sandboxed and actions are approval-gated.

You can override and mark any MCP as trusted, but you accept responsibility for that choice, the same informed-consent pattern we use everywhere else.

No, that doesn't scale. Instead, we require publishers to verify their identity, we display each MCP's requested capabilities at install (like mobile app permissions), and we sandbox execution. Users can report bad MCPs, and flagged ones are automatically demoted pending review. It's a registry, not a curated marketplace.

Yes. Prismr is an open platform. Power users can bring custom APIs, install community MCPs, or script their own tools. You'll see a limitation-of-liability disclaimer when going off the rails, but we don't lock you out of advanced use. Safe defaults stay on for everyone else.

Desktop is the primary product (Mac and Windows today, Linux coming). That's where the full capability lives: running agents, controlling a browser, accessing your local files, and executing MCPs. Monitoring happens in the desktop app, with some actions available via the mobile companion. Mobile (iOS today, Android in January) is lighter: chat, groupchat, approvals, and monitoring. We deliberately don't ship a browser extension.

Not today. The current product uses logical isolation between tenants, which is appropriate for individuals and teams. If your organization has physical-isolation, data-residency, or single-tenant requirements, get in touch. We're happy to discuss dedicated and single-tenant options on a case-by-case basis.

Prismr's audit trail, approval gates, and secret-handling architecture are designed with regulated users in mind, but compliance certifications (SOC 2, HIPAA, etc.) are future work. We're positioned as an informed facilitator: we disclose risk where we can, get your consent, log everything, and let you decide. Regulated-domain use requires your own legal review of whether our posture fits your obligations.