Agent Ops
Agent Ops is Cognipeer Studio's autonomous agent operations module. You create a Project that owns a persistent Linux sandbox, then give it Tasks — one-off, scheduled, recurring, or webhook-triggered — and an AI agent plans and executes each task end to end inside that sandbox: writing and running code, processing files, calling your integrations, browsing the web, and pausing for your approval whenever a decision or a risky action needs a human.
Where a Peer answers in a conversation and a Flow executes a predefined sequence of steps, an Agent Ops task is goal-driven: you describe the outcome, and the agent decides the steps, uses the tools it needs, and delivers the result — with every step streamed live to the task screen.
Key concepts
| Concept | What it is |
|---|---|
| Project | The unit of setup: a system prompt, a model, integrations, skills, environment variables — and one persistent sandbox whose files survive across runs. |
| Task | A unit of work with a title and description. Can run once, at a scheduled time, on a cron schedule, or when a webhook fires. |
| Run | One execution of a task. Every run keeps its plan, logs, tool calls, artifacts, and token usage — and can pause and resume. |
| Sandbox | A real Linux container (Python + Node.js, office document libraries, git) attached to the project. All file work happens here. |
| Action | An integration from the tool gallery (GitHub, Slack, Jira, CRMs, SQL databases, email, and ~60 more) enabled for the project. |
| Skill | A reusable instruction pack — a procedure, format, or policy the agent loads on demand when a task calls for it. |
| Automation | A trigger that creates tasks automatically: a public webhook URL or an integration event such as a GitHub pull request. |
| Template | A ready-made task definition from the gallery (PR review, incident triage, competitor watch, and more). |
What the agent can do
- Execute in a real environment — run shell commands and Python/JavaScript/TypeScript code, install packages, and manage files in the project sandbox. See Execution Environment.
- Produce real documents — Excel workbooks, Word documents, PowerPoint decks, and PDFs, generated programmatically and saved to the project's Files.
- Work with git — clone repositories, commit, and push, with pushes gated behind your approval by default.
- Use your systems — GitHub, Slack, Jira, Notion, CRMs, SQL databases, payment providers, email, and more via Actions.
- Search and read the web — built-in web search and page extraction, plus an optional browser capability for interacting with pages.
- Run on autopilot, pause for humans — schedules and webhooks start tasks without you; approvals and questions stop them when a human decision is needed, with email notifications and deep links back to the task.
- Wait for the outside world — a task can mint a callback URL, hand it to an external system, and suspend for up to seven days until the event arrives. See Automations & Triggers.
- Coordinate work — spawn subtasks, watch other tasks in the project, and wait for them to finish before continuing.
- Learn across runs — important facts, formats, and decisions are saved to project Memory and recalled in later runs.
When to use Agent Ops
| You want… | Use |
|---|---|
| A conversational assistant grounded in your knowledge base | Peers |
| A deterministic, repeatable pipeline with fixed steps | Flows |
| An agent that takes a goal and works autonomously — with files, code, integrations, schedules, and human checkpoints | Agent Ops |
The three compose: an Agent Ops project can be the "operations arm" behind work your team currently does by hand, while Peers handle conversations and Flows handle fixed pipelines.
Section guide
- Getting Started — create your first project and run your first task
- Projects & Sandbox — project setup, runtime settings, environment variables
- Tasks, Runs & Scheduling — task lifecycle, cron schedules, subtasks, templates
- Execution Environment — what's inside the sandbox and how files work
- Skills — teach the agent your procedures and formats
- Actions & Integrations — connect your systems
- Automations & Triggers — webhooks, GitHub triggers, external events
- Approvals & Human-in-the-Loop — approvals, questions, notifications
- Memory — how projects learn over time
- Use Cases — detailed, ready-to-build scenarios by team