Creating an agent

Every configuration option for an Aimogen Pro AI Agent, including model, tool selection, memory, planning, step limits and scheduling.

Aimogen Pro › AI Agents › Manage Agents › Add New

Identity#

Field	What it does
`Agent Description`	What the agent is for. Included in its prompt, so write it as instructions
`Agent Expertise`	Domain framing — the perspective it should work from
`Agent Model`	Which model drives it

Description and expertise are the agent equivalent of a system prompt. Be specific about scope and boundaries:

Description:
You maintain the blog archive of Example Ltd. You find posts with missing
featured images, missing SEO descriptions or broken internal links, and fix
them.

You never change the body text of a post. You never publish, unpublish or
delete anything. If a fix is not clearly correct, you report it instead of
applying it.

Expertise:
WordPress content operations and technical SEO.

Model#

Agents place unusual demands on a model: reliable tool calling, instruction following, and coherence over many steps.

Good choices: claude-sonnet-5, claude-opus-5, gpt-5.5, grok-4.6.

Poor choices: small fast models. They call the wrong tools, loop, and forget constraints.

API level tool calls is automatically disabled when the selected model is not in the function-calling capability list. Text-based tool parsing is a fallback, not an equivalent — pick a tool-capable model.

Tools#

Tools are selected individually, grouped by category, with Allow All Tools as a shortcut.

For the archive-maintenance example above, that is search_site_posts, get_wp_post_data, get_post_structure, manage_wp_post and wp_sideload_media. Nothing else.

Full list in Agent tool reference.

Behaviour#

Setting	Default	What it does
`Max Steps`	—	Maximum reasoning iterations, 1–100. Higher allows complex tasks but increases cost and time
`Max Tool Retries`	—	How many times a failed tool call is retried before giving up, 0–10. Prevents infinite loops on failing tools
`Allow delegation`	off	The agent may assign tasks to other agents
`Enable planning`	off	The agent produces a plan before executing
`Planning required`	off	Forces planning before every action
`API level tool calls`	on where supported	Native model tool calling rather than text parsing
`Short-term memory`	on	Stores recent steps during execution. Required for multi-step reasoning
`Long-term memory`	off	Stores knowledge across runs using embeddings

Max Steps#

The most important setting. It is the ceiling on cost and on how far the agent can wander.

Task	Suggested
Single well-defined action	5–10
Multi-step with a few tools	15–25
Research or audit	30–50
Anything	never 100 on the first run

Planning#

With planning on, the agent writes a plan before acting. That improves performance on complex goals and makes runs much easier to debug, because the plan is recorded.

Planning required forces it every time. Useful for strict workflows; it slows simple tasks down.

Memory#

Short-term memory stores recent steps during the run. Leave it on — without it the agent forgets what it just did.

Long-term memory persists knowledge across runs as embeddings. It requires a configured embeddings provider, model and vector store, and the interface disables the option and says so when they are missing. See Knowledge base and RAG.

Agent memories are namespaced as agents_<agent_id>, and are deleted with the agent.

Delegation#

Allow delegation lets the agent use list_available_ai_agents and delegate_to_ai_agent to hand work to another agent.

Aimogen Pro > AI Agents agent editor

Scheduling#

Schedule Type decides whether the agent runs on demand or on a schedule. Scheduled agents pick up their task automatically. See Running and monitoring agents.

Writing a good task#

The task is the instruction for one run. Good tasks are specific, bounded and verifiable:

Good: "Find the 10 most recently published posts with no featured image. For each, generate a featured image from the post title and attach it. Report the post IDs you changed."

Poor: "Improve the blog."

The good version states scope (10 posts), a criterion (no featured image), an action (generate and attach) and a deliverable (report IDs). The agent knows when it is finished.

Since 2.7.9 you can also write or refine an agent task using AI from the agent screen, which helps turn a vague intention into a specific task.

Import and export#

Download Agents To File exports agent configurations, which can be imported elsewhere. Useful for moving a tested agent from staging to production.

Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.