eesel AI

Network Access connects an agent to external REST APIs using an allowed domain and a stored authentication header. Requests support GET, POST, PATCH and DELETE, including mutations in the customer's own systems. Bearer tokens, basic authentication and custom API key headers are attached on the server without exposing their values to the model. Native helpdesk and commerce connectors also send replies, route tickets and process permitted order changes.

Sourcedocs.eesel.ai/apis-and-developer-resources/network-accessread 2026-10-20

Skills execute jobs across connected applications, such as reading a helpdesk ticket, checking a Shopify order and posting the result to Slack. Customers supply instructions that determine which cases to handle, which actions to take and when to request approval. A skill reports its intermediate steps and saves its result as a file. Custom repeatable jobs can be written into reusable files, while scheduled automations run the same work without a new chat request.

Sourcedocs.eesel.ai/skills/skillsread 2026-10-20

Connected help centers, solved tickets, macros, websites, internal wikis and uploaded files form each agent's searchable knowledge. Past ticket imports can filter by date and status. Helpdesk knowledge updates during operation, while other sources can be refreshed from their integration page or through a chat request. Uploaded PDF, text, Markdown and CSV content needs selectable text. Disconnecting a source removes its associated content from the workspace.

Sourcedocs.eesel.ai/support/faqread 2026-10-20

Each action has an automatic, approval required or disabled setting for each agent. A held action preserves the proposed reply, refund or other change until an Owner or Editor approves or denies it. Approval cards appear in the dashboard and in the originating Slack thread, with terminal controls also available. Always allow changes that action's future permission. Unanswered approvals remain pending and never execute automatically after a timeout.

Sourcedocs.eesel.ai/instructions-and-memory/actions-and-approvalsread 2026-10-20

Workspace access distinguishes Owners, Editors and Viewers, with editing and approval decisions reserved for the first two roles. SAML single sign on and enforced multifactor authentication are paid options. Each agent receives its own source and action permissions, and outbound APIs use domain allowlists with server held credentials.

The SOC 2 Type II audit covering security, availability and confidentiality is in progress, its observation period having started on 1 October 2026. The security program runs on 18 written policies and 50 controls checked by code every morning, with AES 256 encryption at rest, keys in AWS KMS and HIPAA business associate agreements for Enterprise customers.

Sourceeesel.ai/securityread 2026-10-20

Activity records each task across helpdesks, Slack, email and scheduled work. The task view identifies its originating ticket or thread, trigger, searched sources, individual actions, approval decisions and stepwise reasoning. Operators can inspect a failed answer against the content the agent retrieved and the action it selected. Approval records identify who decided and when. Task exports support outside analysis, while Reports aggregates usage, quality and knowledge gaps.

Sourcedocs.eesel.ai/reports/reportsread 2026-10-20

A conversation can continue through its task identifier, retaining context for followup requests. Corrections that a user asks to keep are converted into editable rules on the agent's Instructions page, and those rules apply across channels and later runs. Existing rules are revised when they already cover the point. This persistent material is the agent's operating brief; customer history remains in connected tickets and sources, without a separate customer memory store.

Sourcedocs.eesel.ai/instructions-and-memory/instructions-and-memoryread 2026-10-20

The hosted service runs on AWS, and every plan on the pricing page, the free trial included, lists EU or US data residency. The EU option covers stored customer data and AI processing and is set up with the eesel team. The trust center places full EU residency with Enterprise, so the two pages package the same option differently. Transfers outside the chosen region run under a GDPR data processing agreement with standard contractual clauses. No version of the product runs on a customer's own infrastructure or in its cloud account.

Sourcedocs.eesel.ai/account/security-and-privacyread 2026-10-20

The Skills page offers ready jobs a team starts with one click against its connected data: an onboarding plan, a weekly support report, a Slack answer digest and a knowledge gap report among them. Further shipped skills analyze replies staff rejected, review an agent's setup and draft internal notes on tickets, and each one reports its steps and saves its result as a file. Teams can also write their own repeatable jobs into reusable skill files. Some older workspaces see an installable list instead of report cards, and the knowledge base update skill needs activation by eesel.

Sourcedocs.eesel.ai/skills/skillsread 2026-10-20

Automations wake an agent on customer messages, ticket events, Slack mentions, webhooks or a clock schedule. Customers specify the cases to process in each automation's instructions, and skipped events retain a reason in its activity history. Schedules offer daily, weekly, monthly, hourly and custom timing, with a minimum interval of ten minutes. Zendesk, Freshdesk and Gorgias events can wait up to fifteen minutes for helpdesk rules to finish first.

Sourcedocs.eesel.ai/automations-and-schedules/automations-and-schedulesread 2026-10-20

Scheduled work has a Model strength setting with a regular option for drafting and multistep work and a lighter option for simple tagging or routing. Behind that setting, eesel runs models from OpenAI, Anthropic and Google in its managed runtime and decides which one serves a task. A customer cannot name a model maker, set a routing policy or point the agent at its own inference endpoint. External API credentials configure connected business tools and never replace the underlying language model.

Sourcedocs.eesel.ai/automations-and-schedules/automations-and-schedulesread 2026-10-20

A workspace exposes an HTTP MCP endpoint with bearer authentication and an agent identifier header. The CLI issues a workspace token valid for thirty days and prints the connection configuration. Tools include chat, activity, integrations, automations, instructions and files. Editors can change agent instructions, manage schedules, start source downloads and invite teammates. The token carries the user's role, while mutations continue to obey the workspace's approval settings.

Sourcedocs.eesel.ai/apis-and-developer-resources/claude-code-and-mcp-clientsread 2026-10-20

Helpdesk simulation runs the configured agent against resolved tickets or generated test cases, scores its answers and suggests instruction changes. The report starting point samples twenty recent tickets, compares the generated replies with the team's actual answers and groups results by topic without sending anything to customers. Live chat simulation performs the equivalent replay for closed chats. Followup skills examine rejected or edited replies and propose targeted corrections.

Sourcedocs.eesel.ai/skills/skillsread 2026-10-20

The Chrome extension reads the current page or ticket into a side panel and drafts a response using the selected agent's knowledge. On Zendesk, Paste into Zendesk inserts the draft directly into the reply field; staff can edit it before sending. Other helpdesks use manual copying when that insertion link is absent. The extension can also save selected tabs as knowledge, but does not navigate arbitrary interfaces or operate a general desktop session.

Sourcedocs.eesel.ai/use-cases/browser-extensionread 2026-10-20