Hyperagent
Hyperagent is an agent platform where each named agent gets a real browser, shell and file system, then delivers finished work — sites, decks, documents, dashboards, media — from a thread, a schedule, Slack or a webhook.
What is Hyperagent?
Hyperagent is a platform for building and running AI agents that produce finished work rather than conversation. Published by Hyperagent, Inc. of West Hollywood, California, it rests on two units. The first is the agent: a teammate configured once with a name, instructions, tools, skills, knowledge, a model, its invocations and its level of autonomy. The second is the thread, which holds one piece of work — the conversation, every tool call, and everything produced. Every run becomes a thread, whatever started it.
What separates it from a chat assistant is the environment behind each agent: a real browser, a shell and a file system that persists between turns. Agents search the web through a standard or Exa provider, drive live pages behind logins, execute code, and build deliverables. Those deliverables are the point — functional hosted webpages, versioned decks and documents, typed tables, dashboards, images up to 4K, short video with native audio, speech and multi-speaker dialogue, transcripts with speaker diarization, HeyGen presenter clips, and maps. Everything produced lands in the Library and stays searchable.
A run starts one of four ways: from a thread, from a schedule or Live Mode watching for change, from an external channel — Slack, Telegram, email, a webhook or an MCP client — or by delegation from another agent. Trust is granted per doorway: schedules default to read-only, email addresses take sender allowlists, webhooks want their secret, and connected accounts must be enabled on the agent before it can act with them.
The platform also learns. Skills teach reusable methods, memories hold facts that survive across conversations, and rubrics score whether agents are genuinely improving. Team workspaces share agents, skills and memories under Owner, Editor and Member roles, with one Command Center over the whole fleet.
Models come from several labs — Anthropic, OpenAI and Google alongside open models such as Kimi, GLM, Qwen, DeepSeek and MiniMax — set per agent, overridable per thread, and taken either as a maintained Latest alias or a pinned version for reproducibility.
What it does
- Hand a named agent an assignment and receive a finished deliverable rather than an answer
- Run that agent unattended on a schedule, or in Live Mode watching something and alerting you when it matters
- Trigger it from Slack, Telegram, an email address, a webhook or an MCP client, or by delegation from another agent
- Drive a real browser through live pages, including behind an authenticated session, and hand control back to a human
- Publish hosted webpages, versioned slide decks, documents, typed tables and dashboards straight out of a run
- Generate images, short video with native audio, speech, transcripts with speaker diarization, avatar videos and maps
- Teach reusable skills, save memories and score agent quality against rubrics, shared across a team workspace
When to use Hyperagent / When not to
A quick filter to help you decide if Hyperagent is the right fit.
When to use Hyperagent
- Operations and business teams with a repetitive job worth turning into a named agent that runs on a schedule
- Companies already living inside Slack, a CRM or a data warehouse who want the AI to act in those tools rather than describe them
- Founders and small teams who need research, drafting, design and publishing handled end to end without hiring for each
- Analysts and marketers producing recurring deliverables — briefs, comparisons, dashboards, campaign assets — on a fixed cadence
- Non-technical staff who configure an agent through a form: identity, instructions, tools, knowledge and model, no code required
When not to use Hyperagent
- Anyone with a clinical use case: the terms forbid clinical purposes, substituting for medical advice, and storing protected health information
- Organisations handling special-category personal data, government identifiers, biometrics or ITAR/EAR-controlled material, all barred from input
- Teams needing a documented hosting jurisdiction, a published DPA, an Article 27 EU representative or a security certification
- Users under 18, who are outside the eligibility rules, and anyone unable to accept binding individual arbitration
- Budgets that need a free tier or a fixed monthly ceiling: entry is $20 per month and each run consumes credit at a variable rate
How to use Hyperagent
A typical end-to-end flow, from setup to results.
- Open a thread and ask for a real piece of work; nothing needs to be configured first
- Turn a job worth repeating into a named agent: give it an identity, instructions, tools, knowledge and a model
- Pick the standing tool set on the agent's Tools tab, using the presets, then narrow or widen it per conversation under Settings then Capabilities
- Connect the accounts the work needs under Settings then Integrations, approving access with each vendor
- Enable each connected account on the agent or thread that should use it — connecting alone grants nothing
- Save the first memory or skill from a real run; the agent proposes them, you decide what sticks
- Attach an invocation so work arrives on its own: a schedule, Live Mode, Slack, Telegram, an email address or a webhook
- For unattended runs, decide whether the agent may write to connected apps or only read, and set a budget limit per query
- Point an external MCP client at https://hyperagent.com/api/mcp and complete the browser sign-in; no API key is generated
- Audit results through the Activity tab, where each row opens the thread with the full record of what the agent read, did and produced
Pros & Cons
Pros
- Each agent gets a genuine compute environment rather than a bare model call: real browser, shell and files that survive between turns
- Unusually wide model catalogue across Anthropic, OpenAI, Google and open providers, set per agent and overridable per thread
- Output is a hosted, finished artefact — a site, a deck, a dashboard, a video — not a block of text to paste somewhere else
- Security model is stated plainly: an integration's OAuth token never enters the agent's execution environment
- Trust is scoped per doorway, with restrictive defaults for unattended runs and a per-query budget limit
- Thirteen subprocessors are named in full in the terms, and training on customer input requires express consent
- Documentation is dense, served as Markdown, and published for machine reading through /llms.txt and /llms-full.txt
Cons
- No free tier at all: entry is $20 per month, and Pay As You Go and annual plans were withdrawn on 26 August 2026
- Plan credit expires at the end of each billing period, so anything unspent is lost
- Run cost is not predictable in advance — it depends on the model, the length of the thread and the tools called
- There is no account-level spending cap, only a budget limit per query set on an individual agent
- No hosting country or region is published anywhere, despite thirteen subprocessors being named
- No Article 27 EU representative, no named data protection officer, no published security certification, and the DPA is only referenced conditionally
- The commercial site is minimal — no product, pricing, about, contact or security page; prices live inside the documentation
Pricing & Plans
There is no free plan. Access begins at a paid monthly subscription of $20 (USD), the Hyperagent 20 plan, whose price is charged at the start of each billing period and returned in full as a credit balance the agents spend as they work. Pay As You Go and annual plans were discontinued on 26 August 2026.
- Hyperagent 20 — $20 per month
- $20 of credit
- no bonus
- Hyperagent 50 — $50 per month
- $55 of credit
- 10% bonus
- Hyperagent 100 — $100 per month
- $115 of credit
- 15% bonus
- Hyperagent 200 — $200 per month
- $240 of credit
- 20% bonus
- Hyperagent 500 — $500 per month
- $625 of credit
- 25% bonus
- Hyperagent 1000 — $1
- 000 per month
- $1
- 300 of credit
- 30% bonus
- Hyperagent 2000 — $2
- 000 per month
- $2
- 700 of credit
- 35% bonus
- Hyperagent 5000 — $5
- 000 per month
- $7
- 000 of credit
- 40% bonus
- Hyperagent 10000 — $10
- 000 per month
- $14
- 500 of credit
- 45% bonus
- One-off credit blocks from $10 to $25
- 000
- one dollar of credit per dollar
- valid 90 days and earning no bonus
- Shared Billing moves several members onto one organisation payment method
- it is one-way
- needs a paid web plan
- and excludes Apple-billed subscriptions
Data, GDPR & hosting
A consolidated view of how Hyperagent handles your data.
GDPR overview
Implementation is concrete but partial. Section 10 of the privacy policy, effective 31 August 2026, is dedicated to the EEA, Switzerland and the United Kingdom. Hyperagent, Inc. declares itself the controller of the personal data it holds, names its lawful bases (consent where required, otherwise contract performance or legitimate interest), and grants the seven GDPR rights: access, portability, rectification, erasure, restriction, withdrawal of consent and objection. Rights are exercised by writing to privacy@hyperagent.com with the subject "European Rights Request", and the policy links the EDPB directory for lodging a complaint. Cross-border transfers rely on consent, standard contractual clauses or adequacy decisions. Two gaps stand out: no Article 27 representative is designated, and no data protection officer is named.
Who owns the data?
Under section 3.2 of the terms, prompts and generated results are both "Your Content", and the customer retains all ownership rights in them; the privacy policy repeats that you own what you upload, provided you hold lawful title. Hyperagent may access that content only in a closed list of cases: to provide, maintain, improve or optimise the service, where you explicitly approve access, in response to lawful requests or legal process, to protect system stability and security, and to protect rights or safety. It may also withhold or delete content it believes breaches the terms. Account information and usage data sit outside this protection: they are not Content, and the vendor is free to exploit them for its own business purposes, including after termination.
Reuse rights
You may reuse your prompts and the results freely: ownership stays with you, and no licence back to Hyperagent is claimed over them beyond running the service. Two caveats matter. First, the terms warn that results are not unique — the same or similar output may reach a third party — and that factual assertions should be independently checked before you rely on them. Second, thirteen named subprocessors touch the data in transit: five model providers (Anthropic, Fireworks.ai, Google, OpenAI, Vercel) and eight others including Amazon Web Services, Cloudflare, Browserbase, Exa Labs, HeyGen, Intercom, Mailgun and Serper.dev. Those model providers may retain input and output for up to thirty days for safety and compliance moderation. Training the generative models on your input or output requires your express consent, which the terms otherwise refuse.
Data retention & training
Hosting summary
No hosting jurisdiction is disclosed. Neither the privacy policy, the terms nor the documentation names a country, a region or a data centre where customer data is stored, and no data residency option is offered. What is published is the supply chain rather than its location: Amazon Web Services, Cloudflare and Vercel appear among the named subprocessors as infrastructure providers, and Serper.dev is the only one whose processing location is stated, in the United Kingdom. The privacy policy anticipates transfers outside your own jurisdiction and says it will rely on an appropriate legal mechanism when required — consent, standard contractual clauses or an adequacy decision — and will explain the basis used on request. Security is described in general terms: encryption in transit and at rest, and access to personal data restricted to personnel who need it. No certification such as SOC 2 or ISO 27001 appears anywhere in the published pages. The domain resolves behind Cloudflare, which indicates the delivery network and says nothing about where data actually rests.
Things to keep in mind
Risks and trade-offs to weigh before adopting Hyperagent.
- Spending drifts easily: credit is consumed per model token, image, video second, search, browser minute and integration call, and only a per-query budget caps it
- Unused plan credit expires with the billing period, which quietly rewards over-buying and punishes a slow month
- Auto-recharge is the single setting that authorises charges beyond the plan price, and it is easy to switch on and forget
- An agent on a schedule or in Live Mode acts with nobody watching, so write access to connected apps deserves a deliberate decision rather than a default
- Outsourcing judgement is the real cognitive risk: the terms themselves warn that output may look precise and still be materially wrong, and that human review is expected
- Ownership is asymmetric — you keep your content, but account information and usage data remain exploitable by the vendor even after you leave
- The terms impose individual arbitration and waive class actions unless you opt out in writing to legal@hyperagent.com within thirty days of accepting them
Setup & Integrations
Technical difficulty
Low to moderate. The first run needs no configuration at all: open a thread and ask. Building a named agent is a form — identity, instructions, tools, knowledge, model — with no code, and the official onboarding path is four lessons in about an hour. Most integrations connect in one click; only custom MCP, Slack, HubSpot and Databricks need a dedicated guide. The MCP server requires no API key, just a browser sign-in. The usual stumbling block is procedural rather than technical: connecting an account does not grant it, and the integration must also be enabled on the agent or thread.
Deployment
Integrations
Behind Hyperagent
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Hyperagent.
Frequently asked questions
How much does Hyperagent cost to start?
What happens when my credit runs out?
Which models can an agent use?
Where can an agent be triggered from?
Is my data used to train the models?
Who are the subprocessors?
Does Hyperagent offer an API?
How do I exercise my GDPR rights?
Can a team share agents and knowledge?
Can I migrate an agent from OpenClaw?
Should you pick Hyperagent?
Hyperagent makes a coherent bet: an assistant is worth little unless it can act, so every agent gets a real browser, a shell and a file system, and is judged on the artefact it leaves behind. That artefact is unusually concrete — a hosted page, a versioned deck, a typed table, a dashboard, a video — and it lands in a searchable library rather than in a chat log. Around that core sits an equally deliberate operating model: work arrives through channels a team already uses, trust is granted one doorway at a time, and a learning layer of skills, memories and rubrics is meant to make agents better with use rather than merely busier.
The natural buyer is a team that already has its tools wired together and a repetitive job it can describe precisely. For them the documentation is a genuine asset: dense, honest about what is on by default, and published in a form machines can read.
Two reservations deserve weight. The first is cost predictability. There is no free tier, entry is $20 a month, plan credit expires unspent at the end of each period, and a run's price moves with the model, the thread length and the tools it calls — the only ceiling is a per-query budget on one agent. The second is the compliance perimeter. Thirteen subprocessors are named and the no-training clause is clear, which is more than many rivals offer, but no hosting country is disclosed, no Article 27 representative is designated, no certification is published, and the data processing addendum is only referenced conditionally. Both are questions to settle before committing sensitive work, not reasons to dismiss the platform. Buyers should also weigh how young the company is: nothing is verifiable before 2026.
- Choosing a selection results in a full page refresh.
- Opens in a new window.