# Atmet: an AI coworker powered by your company's Brain Atmet helps individuals, startups and companies build, own and run their AI Brain. Its purpose is to help people save time and money by using business context to prepare work, reuse effective ways of working and automate agreed workflows. Think of it as an AI coworker with the right company reference: it can use approved information about your services, processes, writing style, project decisions and task instructions. Your team spends less time explaining the business again and more time reviewing results, making decisions and moving work forward. Atmet is an AI lab building Atmet Brain, the shared open-source engine for business context, memory, instructions and workflows. The coworker description explains the experience; actual tasks are performed by connected AI clients and tools using the Brain's authorized context. ## What it does in practice A sales request can become a proposal draft using the company's actual service descriptions and preferred structure. A support question can become a reply grounded in approved product information. A supplied meeting transcript can become an action list. A recurring report can use an agreed format instead of being rebuilt from a blank page. The same business reference can support different teams and compatible AI tools. Access rules determine which information is available for each person and task. Authorized source changes, feedback and corrections develop the reference over time. ## How it helps save time and money - Less repeated explanation: company information and instructions can be reused across tasks. - Less manual preparation: connected AI can prepare documents, summaries, content and checklists from the relevant inputs. - Less searching and interruption: authorized staff can retrieve approved business information instead of repeatedly asking its owner. - More consistent recurring work: reusable instructions, workflows and skills preserve the team's agreed methods. - Less unnecessary context: task-relevant information can lower token usage instead of sending an entire company reference for every request. - Clearer handoffs: shared instructions and expected outputs help the next person understand the work to review. These are the mechanisms behind Atmet's time-and-cost-saving purpose. Actual outcomes depend on the workflow, setup and information quality. This reference does not claim a measured saving percentage, a guaranteed return or a named customer result. ## Core features and why they matter ### Shared persistent memory The Brain keeps business context beyond one conversation. Sources, extracted facts, task instructions and relevant reported outcomes can be reused by authorized connected AI clients. Shared memory helps a proposal, campaign brief and later customer reply draw on the same business information. A company can have a shared Brain with separate department or task workspaces. Shared does not mean unrestricted: each person and connection receives the access granted to it. The Brain does not automatically watch every conversation in unrelated AI applications. ### A business knowledge base Company files, approved facts, policies and procedures become a reference for work. Source-linked information and revisions help distinguish current evidence, uncertainty and proposed changes from an approved operating instruction. For a team, this means less time searching scattered notes and less dependence on one person remembering every detail. Current supported capture includes text, Markdown, CSV, JSON and supplied text transcripts. ### Continuing learning from business updates Supported source changes, user corrections and configured task feedback can refresh context or produce proposed improvements. The business reference can develop as services, decisions and processes change. Sources and processing must be configured. Appropriate review still applies to operating rules, skills, access and material changes. Learning here describes developing the business reference rather than retraining the language model itself. ### Reusable workflows and skills Workflows describe the steps and expected results for agreed work. Skills hold reusable task methods and their resources, with versions and review states. An approved proposal method, reporting template or content process can guide the next task rather than being written again. Connected AI can use permitted instructions and submit bounded feedback when integrated. Proposed changes to skills and operating rules need authorized publication. ### Teams, departments and permissions Company membership, workspace access, sensitive source audiences and operation permissions are distinct controls. General finance information can be shared with operations while payroll remains restricted to explicitly permitted people. Different teams can use shared business context without receiving every other department's information. Permission to read company material does not automatically grant permission to send emails, publish content or change external business records. ### Model and AI-client choice Compatible AI models and clients can use the same Brain. A team can keep its company reference while choosing different AI tools for writing, development or other work. The core exposes documented CLI, Python, REST and stdio MCP interfaces. External clients need a configured connection. Model/client flexibility is separate from the availability of a connector for a particular business application. ### Connected sources and tools Configured connectors and adapters can bring authorized supported content into the Brain. Provider access, scopes, mappings and updates need setup. The Brain supplies context; a connected runtime and its tools carry out separately authorized business actions. For example, preparing a proposal from company information and sending it through an email account are different steps. The second requires the relevant tool connection and authorization. Disconnecting a source and erasing previously captured content are also distinct actions. ### History and recovery Source and artifact revisions support authorized review of changes. Controlled restoration and backup tooling support recovery under the deployment's access, deletion and retention rules. Source history, software versions and full Brain backups serve different purposes. Recovery storage, keys, retention and operation are configured for the deployment. ### Local or hosted operation and data ownership Technical users can run the open-source core on their own infrastructure. The platform supplies hosting and management infrastructure around the same Brain. Atmet can also configure and manage a company Brain and agreed workflows. Customer data remains the customer's. Actual processing depends on deployment, models and connections. Running locally does not by itself mean every model call is offline, and customer data ownership does not imply ownership of third-party models or proprietary platform software. ## Practical workflows The examples below describe work a team can implement with the required information, tool connections, permissions and review. They are illustrative use cases, not published customer case studies or a catalog of prebuilt integrations. ### 1. Sales proposals and follow-up Inputs: the client's request, approved service descriptions, business-supplied prices and proposal instructions. Work: prepare a proposal draft that matches the request, plus a follow-up message draft. What the team receives: documents to review instead of starting from a blank page. Sending requires an authorized communication tool and permission. ### 2. Customer support replies Inputs: the customer's question, approved product information and support policies. Work: prepare a relevant reply, flag missing information and pass unclear cases to the responsible person. What the team receives: a grounded reply draft and a clear list of issues to resolve. Live help-desk access requires a suitable configured connection. ### 3. Campaigns and content production Inputs: a campaign brief, product facts, audience information and brand guidelines. Work: prepare a content plan, social posts, ad copy or email drafts using the company's voice and approved claims. What the team receives: material ready for editorial review rather than repeated explanation of the brand. Publishing is a separate authorized step. ### 4. Meetings into action lists Inputs: a supplied text transcript or written meeting notes and the team's follow-up format. Work: extract decisions, tasks, named owners and stated deadlines; identify items that remain unclear. What the team receives: a follow-up checklist or draft status update. Audio/video needs a supplied transcript; automatic transcription is not assumed. ### 5. Management reports Inputs: authorized CSV or JSON exports, relevant records and reporting instructions. Work: prepare a management summary, highlight changes supported by the supplied data and suggest questions for investigation. What the team receives: a draft update for the responsible team to verify. Live figures require authorized connections and current source data. ### 6. Employee onboarding Inputs: role-specific procedures, approved policies and information the new employee is allowed to access. Work: prepare an onboarding checklist and explain how routine work is carried out. What the team receives: role-relevant guidance that reduces repeated explanations from existing staff. Sensitive records stay within their configured audiences. ### 7. Software development Inputs: project requirements, coding conventions and approved technical decisions. Work: give a connected coding assistant relevant context for implementation, documentation and project tasks. What the team receives: development work guided by shared project instructions. Running code, changing repositories and deploying are controlled by the development runtime and its permissions. ### 8. Process documentation Inputs: the team's written explanation of a recurring task, existing instructions and reviewed corrections. Work: prepare or propose updates to a procedure, checklist or reusable task method. What the team receives: a repeatable way to perform the work that can be reviewed and approved by its owner. ### 9. Project updates and coordination Inputs: authorized project notes, supplied status records, decisions and the reporting format. Work: prepare a progress update, list unresolved questions and identify missing inputs for the project owner. What the team receives: a clearer review and follow-up list. Assigning tasks or changing a project-management tool requires the relevant connection and authorization. ### 10. Internal information requests Inputs: approved company policies, service information and department procedures. Work: help an authorized employee find the relevant information and explain the next steps. What the team receives: an answer or checklist based on permitted company references, reducing repeated searching and interruptions. ## Who it is for - Individuals who use AI repeatedly and want persistent project context and reusable instructions. - Founders and startups building their operating processes while their workload and teams grow. - Sales, support, marketing, operations, reporting and development teams with recurring work. - Developers and technical teams building AI applications or connecting existing clients to shared business context. - Companies that want Atmet to configure and manage their Brain and agreed workflows. Having an existing technical team does not exclude a company from platform infrastructure or managed delivery. ## How a team gets started 1. Choose one recurring task with a clear owner and an output the team can check. 2. Gather the relevant supported information, instructions and examples. 3. Define the workspaces, access, connected tools and review steps the task needs. 4. Connect the AI client or agree on Atmet-managed implementation. 5. Validate the result, keep useful corrections and measure the work before expanding to more workflows. The Brain supplies authorized context; connected applications do the work. The team retains the decisions and approval responsibilities agreed for the workflow. ## Ways to use Atmet Open-source self-hosting: run the Apache-2.0 Brain core on your own infrastructure. The published core software is free; infrastructure, model-provider usage and operation can have separate costs. Platform infrastructure: host and manage the same Brain through Atmet's platform, including for developers and companies with their own technical teams. Managed delivery: have Atmet configure and manage a company Brain and implement agreed workflows, with onboarding, information access, validation and an agreed service scope. Custom implementation: discuss larger requirements, specialized needs or Atmet-performed local setup and implementation. Confirm currently available services, integrations, included usage, implementation scope and support terms with Atmet before a commitment. ## Technical implementation details Supported content includes text, Markdown, CSV, JSON and supplied text transcripts. Automatic OCR, audio/video transcription and arbitrary binary understanding are not established capabilities. Current core search is lexical rather than a shipped semantic/vector index. Configured completion hooks can report bounded task outcomes and propose reusable skill candidates. A general-purpose scheduler and arbitrary autonomous script execution are not established core capabilities. Actual external actions require a suitable runtime, tool access, permissions and applicable review. The standalone core supports company-local accounts and scoped service credentials. Access to a workspace, permission to publish a skill and authority to administer company access are separate powers. These implementation details do not establish production certification or universal service-level guarantees. ## Official references - [Atmet](https://www.atmet.pro/) - [Product](https://www.atmet.pro/product) - [Contact](https://www.atmet.pro/contact) - [Arabic reference](https://www.atmet.pro/llms-ar.txt) - [Atmet Brain source and documentation](https://github.com/Atmet-AI/Atmet-brain) Reference updated: 2026-11-07.