HARPA AI
Also known as: Harpa, HARPA AI Technologies LLC, HARPA Grid
HARPA AI is a desktop browser extension that can read pages, fill fields and run sequences of browser actions. A user can ask it to draft a workflow and inspect the steps before running it. It can monitor pages, run commands on a schedule and send results to an inbox. The same extension can use a hosted model, the user's own API key, an existing AI account or a local model.
HARPA AI adds an assistant and automation tools to a desktop browser. The extension can use the current page as context, pull information from an open tab and send it to a model selected for that chat. It can open pages, click elements, enter text into fields and read information from the resulting page. A user can let the model operate in a browser session or ask it to create a reusable command. The command shows its steps in a visual editor, where the user can change which pages it visits, which information it extracts and when it calls a model.
A command can use a loop for a list of pages, check whether a condition applies and stop for a response from the operator. A purchasing team could monitor a list of public supplier pages, collect changed prices and review a digest before sharing it. The extension can run a monitor or scheduled command in tabs while the browser is open. Changes and results appear in an inbox, with browser notifications for monitors. If a scheduled command is missed while the browser is closed, it can run once when the browser opens again. The user can request a manual run to check a configured command before leaving it on a schedule.
The extension supports cloud models from HARPA, including GPT, Claude, Gemini, Grok, Perplexity and DeepSeek, existing AI web sessions, model provider API keys and local models such as those run through Ollama. A chat can use an added MCP server to obtain information or call a tool, and the user can inspect which tools were invoked. Developers can also use the Grid API, served from Frankfurt, to dispatch browser work to configured nodes or connect workflow results to external tools.
Team spaces share commands, connections and a cloud credit balance, with the owner controlling connections and billing and locked commands that members can run but not edit. Memory can preserve selected information across chats, and paid plans can synchronize memories and chats across devices. A user can turn memory off for a chat or edit what the extension remembers.
The free plan offers 3 commands and 3 automations, 100 chats a day on an existing AI login and a small cloud allowance to try. In the United States, Pro costs $12 a month billed monthly or $144 a year with extra credits, and adds cloud model credits, synchronization and unlimited commands. Pro+ at $19 a month triples the credits, Teams starts at $12 per seat and a one time Lifetime plan costs $240. Cloud usage depends on the model and each request's size, and top up packs never expire.
People using an external login or their own provider key avoid consuming HARPA credits for those chats. Page content goes to the selected provider when the user or an automation includes it in a request. HARPA AI Technologies LLC of Helsinki, Finland operates the product. The extension fits individual operators and teams with recurring browser work, such as collecting information from approved web sources, checking product listings for changes or drafting a response from information already open in a tab.
Vendor details
Canonical URL
https://harpa.ai/
Category
Browser / computer-use agent
Subcategory
Browser extension for page aware AI, browser automation and monitoring
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
HTTP MCP servers, Make.com webhooks and connectors, HARPA Grid REST API, external model accounts and API keys.
In practice
A team monitors public product pages and approves a summary before forwarding it.
An analyst extracts recurring values from approved pages and reads results in the automation inbox.
An operator changes the model for a single chat and reviews the MCP tool calls used in the answer.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
13.0 / 14 capabilities · 93%
| Integrations & Tool Calling | Full |
|---|---|
|
HARPA calls tools from connected MCP servers and can post automation results to external services through webhooks. A command can combine browser actions with a request to another service. The Grid API allows a developer to start work on a configured browser node and retrieve the result. A team can also pass output to Slack or a spreadsheet through a Make.com workflow. Sourceharpa.ai/guides/automation-featuresread 2026-11-20 |
|
| Workflow Orchestration | Full |
|
A command combines browser navigation, extraction, model calls and messages as steps on a flow canvas. Groups and loops can process lists, while conditions and jumps change the order of execution. HARPA can build a command from instructions and show the steps for a person to edit. The resulting command can run again using the same sequence and the inputs supplied for that run. Sourceharpa.ai/chatml/automation-stepsread 2026-11-20 |
|
| Knowledge Grounding & RAG | Full |
|
The extension can read the current page, use open tabs and bring page text into an answer. A compatible MCP server can supply additional information from a private document source or database when the user connects it. The model can pull results through those tools in the same conversation. Page access can be enabled separately for each chat, and users can see the tools that returned context. Sourceharpa.ai/guides/mcp-connectionsread 2026-11-20 |
|
| Human Oversight & Guardrails | Full |
|
A workflow can include a question that pauses the run until a person answers. The automation inbox keeps the pending item under Needs you. An operator can use the gate before handing the result to another service. The chat also has separate controls for whether the model may act on the browser, read page content, run commands, search the web or use memory. Sourceharpa.ai/guides/schedule-ai-commandsread 2026-11-20 |
|
| Security, Identity & Governance | Partial |
|
The extension gives users controls for page access, memory synchronization and remote browser jobs. In a team space the owner alone changes shared connections, billing and the plan, members can run but not edit locked commands, and only the space admin can create the Grid API key, which is shown once and can be revoked. With a user's own API key or login, prompts go straight from the browser to the provider, while HARPA's cloud models route prompts through its servers. HARPA handles personal data under the GDPR and claims no SOC 2 or ISO certification. Sourceharpa.ai/privacyread 2026-11-20 |
|
| Observability & Auditability | Full |
|
Each scheduled run can show its last result and next run in the automations workspace, and completed responses appear in the inbox. An approval request remains visible until someone answers it. The user can open the steps that ran for a command and see which MCP tools the model used, including the inputs sent to each tool when that detail is available. Sourceharpa.ai/guides/schedule-ai-commandsread 2026-11-20 |
|
| Memory & State Persistence | Full |
|
HARPA keeps memories between chats when the feature is active. The memory screen lets a user review, edit or delete entries and allows memory to be paused for an individual chat. Paid accounts can synchronize chat and memory with other devices, and that synchronization can be switched off. The app also preserves commands and automation settings for repeat use. Sourceharpa.ai/blog/harpa-changelogread 2026-11-20 |
|
| Deployment & Data Residency | Full |
|
Operators can run the extension in their own desktop browser and connect an inference server running on their machine. Local schedules and page monitors operate in that browser while it is open. The product can also connect to HARPA's hosted models or delegate work through Grid to configured browser nodes. Grid can use a participating browser node that holds the session needed for the task. Sourceharpa.ai/guides/getting-startedread 2026-11-20 |
|
| Prebuilt Agents / Templates / Packs | Full |
|
The browser extension includes a library of supplied commands for work such as summarizing pages, drafting email replies and extracting details from web content. Users can start with a built in command or create their own as a set of steps. Paid plans can edit supplied commands, and Teams can share a selected command and its underlying workflow with other members of a shared space. Sourceharpa.ai/libraryread 2026-11-20 |
|
| Triggers & Channel Coverage | Full |
|
Scheduled commands run at set times and can also start when a user opens a site or another automation finishes. A page monitor can start a command after noticing a change. On a local browser the schedule runs while the browser is open, and a missed run can resume once when it reopens. The result and next scheduled run appear in the automation workspace. Sourceharpa.ai/guides/schedule-ai-commandsread 2026-11-20 |
|
| Model Flexibility & Routing | Full |
|
A user can choose a hosted model for each chat or connect through a supported AI login, an external provider key or a local model. The cloud service offers models from several providers including OpenAI, Anthropic and Google. The selection can change from one conversation to another. A model choice also determines whether the run consumes credits from HARPA or uses the user's separate account. Sourceharpa.ai/guides/getting-startedread 2026-11-20 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
The extension can add MCP servers over HTTP and make tools from those connections available in compatible model sessions. Developers can also call the Grid REST API to dispatch a browser job from an external system and receive its result. A custom command can use a request step to pass an output to a webhook, which can then feed a separate tool or work queue. Sourceharpa.ai/guides/mcp-connectionsread 2026-11-20 |
|
| Testing, Debugging & Optimization | Partial |
|
A developer or operator can open the steps in a workflow, run it immediately and look at the result before relying on a scheduled execution. The MCP connection setup tests a server address and shows the tool count or an error. These checks help debug an individual workflow. The software has no evaluation suite for running a set of labeled tasks and comparing model outcomes over time. Sourceharpa.ai/guides/schedule-ai-commandsread 2026-11-20 |
|
| Browser / Computer-use | Full |
|
The extension can work in browser tabs, including navigating to pages, clicking elements, entering text and inspecting the result. A repeatable workflow can use those actions along with conditions and extraction in the underlying browser session. An operator can keep the page action steps visible in the editor and allow a command to wait for a response before advancing. Sourceharpa.airead 2026-11-20 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
$12 per month, billed monthly or at $144 a year
Subscription per person and cloud model credits
Included quota
Pro includes about 150 cloud requests a month, or about 2,300 a year on yearly billing. Pro+ at $19 a month includes about 450 a month or 7,000 a year. Free includes about 15 cloud requests to try, 100 chats a day on the user's own AI login and 1,000 page aware runs.
What is public
Free includes 3 custom commands and 3 automations. In the United States, Pro costs $12 a month billed monthly, or $144 billed yearly with 25% more credits, and Pro+ costs $19 a month or $228 a year with three times the credits. Teams starts at $12 per seat per month, and a one time $240 Lifetime plan covers one personal seat.
Billing mechanics
Cloud model requests draw credits according to model and request size. A separate AI login or API key does not consume HARPA credits. Additional credit packs are available.
Additional watchouts
Switching between monthly and yearly billing means cancelling and subscribing again. A cloud request counts as about one page and a prompt, and larger models use more credits per request.
Overage / add-ons
Top up packs are paid once, never expire and work on every plan, from about $8 for 150 requests to about $670 for 22,000.
Sales call required
No, self serve available
Free / trial
Free with limited commands and automations
Lowest paid plan
Pro
Commercial notes
Teams can pool the credit allowance and share commands and connections.
Missing data
Large team and custom system quotes depend on the scope of the rollout.
Related vendors
- AgentQL — A self healing query language for the web, with natural language…
- AGI, Inc. — AGI, Inc
- Airtop — Cloud browsers for AI agents that compile plain English workflows…
- Asteroid — Healthcare portal integration platform whose supervised browser and…
- Automat AI — AI agents that operate computers visually, sold as Automat Workforce…
- Autotab — General AI agent that learns a workflow from a demonstration and…
Alternatives to HARPA AI
Other products that do the work HARPA AI does, and what each one is.
- OpenClaw
- Automat AI
- Browser UseOpen source browser agent library
- NaradaAgentic process automation and computer use platform
- SkyvernOpen source browser agent
- Axiom.aiNo code browser automation tool and Chrome extension for building bots