Create and manage agents
Give a recurring assistant its own instructions, files, tools and triggers.
Create an agent
- Open Agents and create an agent with a recognizable name.
- Choose an allowed model and describe its role, expected output and approval boundaries.
- Add only the files, skills and plugins needed for its work.
- Save the settings and send a small first task in its chat.
- Check the work log, sources and generated result before assigning larger or recurring work.
An agent is already task-oriented. It does not need the automatic chat/agent selection used by normal chats.
Instructions, files and memory
Use instructions for enduring behavior, files for reference material and agent memory for retained working knowledge. Keep instructions concise enough to distinguish rules from background information. A connected tool still uses the relevant permissions; adding it to an agent does not turn the agent into an administrator.
Background jobs and triggers
Use the agent's trigger/job settings for scheduled or event-based work. See trigger types. Pausing a trigger prevents future starts; it does not necessarily stop an already running job. Stop the active job separately when necessary.
Each automated job has its own run details, actions and results. The main conversation can show a compact result card. Open that card to inspect the job rather than assuming that its output belongs to the manual conversation's current task.
Concurrent conversations and jobs
A scheduled job can finish while you are working in the agent's main chat. Use the job's start time, trigger and run details to identify its result. A job result card is separate from the answer to your current message. When a result seems missing, check job details before triggering the same action again.
Files, statistics and checkpoints
The agent's settings expose file/storage information where available. Use checkpoints to preserve a supported restorable state. Pause the agent and finish or stop active jobs before restoring. A checkpoint is not an undo for an email, website change or other action already performed outside CogLake.
Deleting an agent
Read the confirmation carefully. Deleting an agent also removes its associated managed data and files; download anything you still need first. It does not reverse actions previously performed in external services. Remove or redirect dependent automations before deletion.
For unavailable or failed jobs, see task states and recovery.