Gobii
Gobii runs persistent AI employees for teams: agents that browse the web, gather data, fill in forms and report back inside the tools you already use. Its final day of operations is September 1, 2026.
What is Gobii?
Gobii is an AI agent platform from Gobii Inc., a San Francisco company launched in June 2025 and backed by Open Core Ventures. Its pitch is deliberately not chatbot-shaped: AI employees for teams with real work to do. An agent here is a durable resource with its own identity, configuration, schedule, messages, files and observable work. You brief it once with a charter, and it keeps going in the background.
Those agents browse the web through cloud-hosted browsers, gather research, organise data, fill in forms, operate connected applications and hand back reports. You reach them by chat, by email or by SMS, which is how non-technical colleagues use the product without touching code. Around that sit persistent memory, built-in tools, schema-guided structured output, live previews, audit trails and execution guardrails. An agent can pause to ask a human for a secret, a contact or an approval, and daily credit caps bound recurring work.
Gobii for Teams turns this into a shared workspace: shared agents and otherwise private templates, shared credentials, API keys, secrets and MCP servers, roles and invitations, seats, a single invoice and pooled credits. Developers get four surfaces, namely a REST Agent API authenticated with an X-Api-Key header, remote MCP, webhooks and events, and a legacy Browser Task API. The core is open source under the MIT licence on GitHub, Kubernetes-native, and deployable on-premise, in the cloud or fully air-gapped with your own models; the platform was originally built on the open-source browser-use framework. Recruiting became the flagship use case, summed up on the site as recruiting agents that deliver qualified candidates into the tools you already use. Third-party traction was real, with number-one rankings on OpenRouter for DeepSeek V3.2 in December 2025, for Mistral Large 3, and worldwide for GLM-4.6 in November.
None of it stays available. Gobii is winding down, with a final day of operations on September 1, 2026. The company has prepared a ZIP export tool for agent data along with instructions for moving to Manus, Hermes, ChatGPT, Gemini and other platforms, and says refunds for active paid subscriptions are being reviewed.
What it does
- Create a persistent agent and give it a charter instead of a throwaway prompt
- Schedule its work with a cron expression, or trigger it by message, webhook or email
- Give it cloud browsers, files, spreadsheets, connected apps and MCP servers to work with
- Inspect the agent timeline: messages, task markers, processing state, files and deliverables
- Share agents, templates, credentials and pooled credits inside a team workspace
- Export results to Google Sheets, CSV, webhooks or an applicant tracking system
When to use Gobii / When not to
A quick filter to help you decide if Gobii is the right fit.
When to use Gobii
- Recruiting teams that want a job brief turned into a sourced shortlist and handed back into LinkedIn Recruiter, Greenhouse or Google Sheets
- Sales and business development teams running account research, buying-signal monitoring and personalised outreach preparation
- Engineering teams that would rather create durable agents through a REST API, cron schedules, webhooks and remote MCP than click through a console
- Regulated organisations that need auditability and data locality, since the core can be deployed on-premise, in the cloud or fully air-gapped
- Operations, competitive intelligence and compliance monitoring roles buried in repetitive but consequential web work
When not to use Gobii
- Anyone starting a new automation project today, since Gobii's final day of operations is September 1, 2026
- Buyers looking for a permanently free tool: the published grid starts at 50 USD per month once the 7-day trial ends
- Teams that want a generic template directory or a chatbot-only interface, which Gobii explicitly says it is not
- Recruiters expecting a replacement for recruiter judgment or hiring compliance review, which the vendor rules out itself
- Users who need a mobile app, a browser extension or an interface in a language other than English, none of which is offered
How to use Gobii
A typical end-to-end flow, from setup to results.
- Create an account from the sign-up page, or log in if you already have one
- Start from a public template in the agent library, such as the B2B Lead Research or Candidate Sourcing AI Employee
- Customise the agent's charter, its tools, its schedule and the output format you expect
- Or take the API route: get a key and POST to https://gobii.ai/api/v1/agents/ with an X-Api-Key header and a JSON body carrying name, charter and schedule
- Read the response, which returns id, name, schedule, is_active, life_state and created_at
- Put the agent to work on a cron schedule, send it a message, or fire an inbound webhook
- Follow the timeline for messages, task markers, processing state, files and deliverables
- Activate or deactivate an agent when you need to pause it, without deleting it
- For a team, create the workspace in the app, name it, invite colleagues and assign roles
- To self-host, follow the quickstart at docs.gobii.ai/self-hosted/overview and the GitHub repository, with the API reference at docs.gobii.ai/api-reference
Pros & Cons
Pros
- The core really is open source under the MIT licence and self-hostable, with your own models
- On-premise, cloud or fully air-gapped deployment, which is rare in this segment
- Durable agents with memory, a schedule and a timeline, rather than throwaway runs
- A rich, documented developer surface: REST, remote MCP, webhooks and structured output
- No-code access by email and SMS for colleagues who will never open a console
- Concrete guardrails, including human approvals, scoped keys, per-agent tool access and credit caps
- A public, readable pricing grid with per-task overage at 0.10 then 0.04 USD, and unusually detailed legal pages for a company of this size
Cons
- The service shuts down on September 1, 2026, which is disqualifying for any new deployment
- No permanently free plan on the published grid: entry is 50 USD per month once the 7-day trial ends, and trials are reserved for new customers, once per person
- An unresolved contradiction: the terms, updated May 29, 2026, mention a free tier of service that the pricing page never shows
- No data processing agreement published or offered, and no named list of subprocessors, only categories of vendors
- No security certification claimed: no SOC 2, no ISO 27001, no HIPAA
- Model tuning appears among the declared uses of data with no documented opt-out, and no Article 27 EU representative is designated even though the service addresses people in the EEA
- Hosting is primarily in the United States with no data residency option on the cloud offer, support runs mainly through Discord and forms, and there is no declared interface language, no mobile app and no browser extension
Pricing & Plans
No permanently free plan appears on the published pricing grid. The entry point is the Pro plan at 50.00 USD per month, which includes 1,000 tasks and then charges 0.10 USD per additional task, with 20 contacts per agent and 600 API requests per minute. Scale costs 250.00 USD per month for 10,000 tasks, then 0.04 USD per task, with 50 contacts per agent, 1,500 requests per minute, priority queueing and a dedicated channel. Team is priced per seat, at 50.00 USD per seat per month, with 1,000 task credits per seat pooled across the workspace, 2,000 requests per minute and usage-based overage. Enterprise is quoted on request, with dedicated infrastructure, an SLA and tailored governance. A 7-day free trial is available on Pro and Scale with no immediate charge and cancellation at any time, and Pro and Scale are billed monthly with no commitment. The MIT-licensed core carries no licence fee if self-hosted, leaving only infrastructure costs. All amounts are published in US dollars, with no currency selector and no regional pricing announced. Gobii defines a task as a single automation job submitted to the platform, each submission counting as one task against the quota. These prices describe a service whose operations end on September 1, 2026.
- 50 USD per month
- 250 USD per month
- 50 USD per seat per month
- pricing on request
Data, GDPR & hosting
A consolidated view of how Gobii handles your data.
GDPR overview
Gobii's privacy policy is written under U.S. law and, in its own words, designed to satisfy key obligations under the GDPR, the California Consumer Privacy Rights Act and emerging 2025 state privacy laws. That is a careful formulation rather than a claim of full compliance. Concrete machinery is nonetheless in place: rights of access, rectification, erasure, portability, restriction and objection, withdrawal of consent, and the right to lodge a complaint with a supervisory authority; named legal bases; standard contractual clauses or another approved mechanism for people in the EEA or the UK, whose data is hosted primarily in the United States; and a dedicated deletion page promising acknowledgement within five business days and erasure within 30 days. Requests go to privacy@gobii.com or through dashboard controls, after identity verification. What is missing is equally concrete: no Article 27 EU representative is designated, and no data protection officer is named.
Who owns the data?
Gobii Inc. publishes and operates the service, but you keep what you put into it. The terms state that you retain ownership of your own content submitted to the Service. The privacy policy goes further on scraped material: Gobii says it does not intentionally ingest personal information that your agents scrape unless you direct the agent to send it back to its servers, and that you are the controller of that downstream data. That splits responsibility cleanly. Gobii processes account, billing and automation-log data as the operator, while the personal data your agents collect from third-party sites stays yours to justify, secure and delete.
Reuse rights
Because you keep ownership of the content you submit and of what your agents collect, you can reuse those outputs freely, exporting them to Google Sheets, CSV, a webhook or an applicant tracking system without asking Gobii for permission. The counterpart is that the lawfulness of how the data was gathered on third-party sites remains your problem. On its own side, the privacy policy lists five purposes: providing and operating the service; security, fraud, abuse and audit work; communication; legal compliance; and improving and developing new features through aggregated analytics, error logging and model tuning. That last phrase is the ambiguity worth flagging, because model tuning is named as a purpose while the policy never says whether customer data feeds training, nor how to object. Data is shared with contractually bound vendors (cloud hosting, payment, analytics, support), with the third-party sites your agents are instructed to visit, with authorities, in a corporate transfer, and with your consent. Gobii states that it does not sell or share personal information for cross-context behavioural advertising as the CPRA defines those terms. The GDPR legal bases cited are contract performance, legitimate interest, consent and legal obligation.
Data retention & training
Hosting summary
Gobii states that it hosts data primarily in the United States. For people located in the EEA, the UK or another transfer-restricted region, it relies on standard contractual clauses or another legally approved mechanism. The announced security measures include encryption in transit, segmented VPC networks, least-privilege access, regular penetration testing and documented incident response procedures, and the company's own launch communication adds a Kubernetes-native architecture, encrypted at rest, with strict access controls. The public site is served behind Cloudflare, so the IP address that resolves is anycast and says nothing about where data actually lives. Two limits matter. No data residency option is offered on the cloud service, so United States hosting is the only choice there. Self-hosting, on the other hand, moves hosting entirely to the customer, air-gapped deployments included, and is the only way to place the data under another jurisdiction. With operations ending on September 1, 2026, anyone holding data in the hosted service should plan its export and deletion rather than rely on these arrangements.
Things to keep in mind
Risks and trade-offs to weigh before adopting Gobii.
- Operations end on September 1, 2026: check that the announced ZIP export tool actually works before depending on anything still stored in Gobii
- Active paid subscriptions are handled by email, with next steps and possible refunds, but no public refund scale has been published
- Free trials convert automatically into a paid subscription unless you cancel before they end, and the terms advertise a free tier the pricing page never shows
- Liability is capped very low, at the greater of the fees paid over the previous six months or 100 USD, under Delaware law with exclusive jurisdiction in Delaware courts
- No subprocessor list is published, so the processing chain cannot be mapped; the postal address is a PMB mailbox rather than an occupied head office; and the privacy contact sits on gobii.com, a different top-level domain from the site, which is worth verifying before sending any formal notice
- Supported regions are never published, letting an agent act on third-party sites remains your responsibility under the terms, and the SMS features forbid cold outreach and purchased lists
- Delegating judgment is the human risk: an always-on agent that files sourced shortlists and research reports invites the habit of accepting its output unchecked, and the vendor itself warns that it does not replace recruiter judgment or hiring compliance review
Setup & Integrations
Technical difficulty
Difficulty depends entirely on the route taken. The product route is not technical: create an account, pick a public agent template and rewrite its charter, with no code involved, after which non-technical colleagues can interact by email or SMS. The API route is easy for a developer, needing one key and one JSON POST with an X-Api-Key header, covered by a single-command quickstart. Setting up a team workspace is light administration: create it, invite people, assign roles. Self-hosting is markedly harder, requiring a Kubernetes deployment and your own model access.
Deployment
Integrations
Behind Gobii
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Gobii.
Frequently asked questions
Is Gobii still available?
What counts as a task?
How does the pricing work?
Is there a commitment?
Is there an enterprise offer?
Can Gobii be self-hosted?
Is there an API?
Where is the data hosted?
What is the minimum age?
How do I get my data deleted?
Should you pick Gobii?
Gobii is a technically serious product arriving at the wrong ending. The core is genuinely open source under the MIT licence, agents are durable rather than disposable, the developer surface is documented across REST, remote MCP and webhooks, the execution guardrails are concrete, and deployment can go as far as fully air-gapped, a combination few competitors in browser automation offer. Traction was measurable by third parties too, with number-one OpenRouter rankings on three different models in late 2025.
None of that changes the verdict. Gobii Inc. is winding down and its final day of operations is September 1, 2026. No new project can reasonably be built on the hosted service, and the honest reading of the pricing, teams and solutions pages still online is that they describe a product about to be switched off. The vendor has at least planned an exit: a ZIP export of agent data, instructions for migrating to Manus, Hermes, ChatGPT, Gemini and other platforms, and refunds for active paid subscriptions said to be under review.
The one form of continuity is the code. Because the core stays available under the MIT licence, a team with Kubernetes skills and its own model access can keep the platform running after the cloud offer goes dark. That is the only recommendation this listing can honestly make today.
The blind spots will not be fixed now, and they matter on the way out: no data processing agreement, no named subprocessors, no security certification, no Article 27 EU representative, and no published retention rule for hosted data after September 1, 2026. Anyone with data still inside Gobii should export it and file a deletion request rather than wait and see.
- Choosing a selection results in a full page refresh.
- Opens in a new window.