Agents
AI Agent
An AI agent that can call tools to accomplish tasks, and with Deep agent on also plans, keeps files and delegates to a sub-agent
Category: Agents · Plans: Trial, Starter, Pro, Premium, Enterprise · Type: AIAgent
Inputs#
| Field | Type | Required | Default | What it does |
|---|---|---|---|---|
| LLM | Choice | No | default | The language model to use for the agent. A workspace model runs its primary model first and falls back down its chain on a provider outage. ‘default’ resolves to the workspace’s marked default model and falls back to gpt-5.4 when none is set. Options: default, flowhunt-recommended, flowhunt-cheapest, flowhunt-smartest, gpt-4o-mini, gpt-4.1, gpt-5, gpt-5-mini, gpt-5-nano, gpt-5.2, gpt-5.4, gpt-5.4-mini and 59 more |
| Reasoning Effort | Text | No | none | Reasoning effort for reasoning-capable models (OpenAI gpt-5 / o-series, Anthropic 4.6+ and the models served through OpenRouter that can reason). The available values depend on the selected model. Options: none, low, medium, high, xhigh, max |
| Deep agent | Yes/No | No | Off | Gives the agent a file system, a planning tool and one helper sub-agent it can hand work to. Use it for long, multi-step tasks; a plain agent answers faster. |
| Temperature | Number | No | 0.7 | Sampling temperature. Higher values make output more random, lower values more deterministic. The valid range depends on the selected model (e.g. Anthropic models cap at 1.0). Hidden for reasoning models, which ignore it. |
| Max Output Tokens | Number | No | 3000 | Maximum tokens in the response. |
| Input | Text | No | {input} | The input prompt to pass to the agent. Use {input} for connected step output. |
| Tools | tools | No | — | Configure tools available to the agent for task completion. |
| System Message | Text | No | You are a helpful assistant. | System message defining the agent’s behavior and role. |
| Max Iterations | Number | No | 20 | Maximum number of iterations the agent can perform. |
| Include Chat History | Yes/No | No | On | Include conversation history in the agent context |
| History Message Count | Number | No | 10 | Number of recent messages to include |
| History Max Tokens | Number | No | 8000 | Maximum tokens for chat history |
| Message Types | Text | No | User Messages, AI Messages | Select which message types to include in chat history Options: User Messages, AI Messages, Tool Calls |
| Show Progress | Yes/No | No | On | Show agent progress and tool calls in real-time. |
| Enable Memory | Yes/No | No | Off | Enable persistent memory for the agent to store and retrieve information. |
| Access Mode | Text | No | Read Only | Control memory access permissions. Read Only allows searching, Read/Write allows adding and updating. Options: Read Only, Read/Write |
| Memory Category | Choice | Yes | — | The memory category to use for storing and retrieving memories. |
| Read Memory Prompt | Text | No | You can use the memory tools to recall persistent context … | Instructions for the Agent how and when to use the Read memory tools. |
| Write Memory Prompt | Text | No | Use add_memory / update_memory ONLY when one of these is … | Instructions for the Agent how and when to use the Write memory tools. Only visible in Read/Write mode. |
| Offload Flow Input | Yes/No | No | Off | When enabled, the flow input will be stored in block storage and can be retrieved later using the get_offloaded_memory tool. |
| Structured Output | Yes/No | No | Off | Enable structured JSON output with a defined schema |
| Output Schema | table | No | — | Define output fields: name, type (string/int/float/bool/list/dict), description |
Outputs#
| Output | Type |
|---|---|
| Message | Any |
| Iterations | Number |
| Tools Used | List |
| Chat History Enabled | Yes/No |
| Memory Enabled | Yes/No |
| Structured Output Enabled | Any |
| Reasoning Effort | Any |
| Artefacts | List |
Notes#
The AI Agent calls the tools you connect to its Tools input to complete a task, and returns its reply as the Message output. Turn on Deep agent to give it a file system, a planning tool and one helper sub-agent it can hand work to, which suits long, multi-step tasks; a plain agent answers faster. Chat history, memory and structured output are each switched on separately.