AZAll System Prompt

The node that makes an AgentAZAll agent feel like an agent instead of a chat endpoint with a memory add-on. AgentAZAll_SystemPrompt assembles everything the LLM needs to know - identity, current task, recent memories, inbox summary, optional notes, and the tool-calling instructions - into a single system prompt string. You wire its output into the LLM node's system_prompt input and suddenly the model knows who it is, what it's been doing, and what it's allowed to do, all from files it would never have seen on its own.

It's the integration hub of the pack: one node, one STRING output, and most of the other nodes quietly feed it without you wiring anything. The demos lean on this constantly - the Character Design Studio's "art director persona" is a System Prompt with a base prompt plus the auto-assembled context.

How it works

Give it a config and it reads the agent's current identity (from Who Am I), current task (from Doing), the N most recent memories (from the Remember store), today's inbox summary (from the mailbox), and optionally your notes. It stacks them into labeled ## sections - "Your Identity", "Current Task", "Your Memories", "Inbox", "Notes" - then appends a built-in "Available Tools" section that tells the model how to emit [TOOL:...] calls for the Tool Parser. Your base_prompt goes at the top, so you can layer a persona over the auto-context.

The memory section respects max_memory_entries, and each memory is truncated to a 200-character preview - so it's a digest, not a full dump. That's a deliberate design choice to keep the prompt from bloating, but it means long memories arrive truncated.

The inputs that matter

• config (required) - from AZAll Setup.
• base_prompt - your own persona/instructions, prepended to everything. "You are an exacting art director who values visual coherence across characters…" - this is where your actual personality for the agent lives.
• include_memories (default true) - fold in the memory digest.
• include_inbox (default true) - fold in today's message summary.
• include_notes (default false) - fold in your notes; off by default because notes can be long.
• max_memory_entries (default 20, 0–100) - how many memories to include; 0 disables them.

The system_prompt output goes straight into the LLM node's system_prompt input. That's the whole wiring story.

Installing it

In cronos3k/comfyui-agentazall - Manager search "AgentAZAll", or:

cd ComfyUI/custom_nodes
git clone https://github.com/cronos3k/comfyui-agentazall.git
pip install "agentazall>=1.0.13"
# restart ComfyUI

No models, no keys.

Common issues

The main thing people hit is prompt bloat: with include_notes on and max_memory_entries high, the assembled prompt can get long enough to eat context - trim the knobs rather than the workflow. Also remember the memory previews are truncated, so if the LLM needs the full text of a memory, the digest won't give it; pair this with Recall for full content. And because it reads state that other nodes write, ordering matters - System Prompt reflects whatever's on disk when it runs, so run the writes (Remember, Send, Doing) before it in the graph if you want this run's changes included.