{% extends "base.html" %} {% block title %}{{ form_title }} - Memorizz{% endblock %} {% block content %}
Configure how this agent behaves and runs.
{{ agent_id }}
Maximum tool-calling rounds per run (default: 20).
Continuous mode may consume a very large number of tokens. The agent will keep running tool loops until the task is complete.
Default http://127.0.0.1:8080/v1 for llama-server. LM Studio uses http://127.0.0.1:1234/v1. The API key is ignored by these servers.
http://127.0.0.1:8080/v1
llama-server
http://127.0.0.1:1234/v1
When on, Gemma 4 / DeepSeek-R1 / Qwen3-thinking emit a separate reasoning trace before the final answer (shown in the playground's trace timeline). Off by default for snappier replies — thinking can take 30–60 s before any visible content on CPU-only setups.
Optional provider-specific fields (e.g. azure_endpoint, api_version, ollama host, HuggingFace settings, temperature, base_url for local-openai).
Presets apply the role's default goals and background. Edit the fields below afterwards to customize.
Selecting a saved persona links this agent to its evolution history. Edits become a new version when saved.
Comma or newline separated memory IDs.
Enable code execution in a secure sandbox. Requires the corresponding API key in Settings (E2B, Daytona) or GraalPy installed locally.
When enabled, the agent gets `internet_search` and `open_web_page` tools. Configure API keys in Settings.
When enabled (and Oracle is connected), the agent can create durable scheduled jobs from chat.
IANA timezone name used when timezone is omitted for automations.
When enabled, this agent can be set as the active WhatsApp communicator. Users can then chat with this agent via WhatsApp.
This message will be sent when a user first contacts the agent via WhatsApp (feature coming soon).
When enabled, the agent can inspect files and run allowlisted commands within configured roots.
One path per line (or comma-separated). Empty uses the current working directory.
Disabled by default. Enables file creation and edits under allowed roots.
Deletes require writes and remain guarded by root and wildcard protections.
{% if skills_marketplace_provider == 'vercel' %} Agent gets vercel_skills_search and vercel_skill_fetch tools to search and use skills from the open Vercel ecosystem. A GITHUB_TOKEN is required for search (raises rate limits from 60 to 5,000/hr). {% elif skills_marketplace_provider == 'skillsmp' %} Agent gets the skills_marketplace_search tool. Configure SKILLSMP_API_KEY in Settings. {% else %} When enabled, the agent gets marketplace search tools. SkillsMP requires SKILLSMP_API_KEY. Vercel Skills requires GITHUB_TOKEN for search. {% endif %}
vercel_skills_search
vercel_skill_fetch
GITHUB_TOKEN
skills_marketplace_search
SKILLSMP_API_KEY
Vercel Agent Skills is selected but GITHUB_TOKEN is not set. Skill search will fail (HTTP 401) until you add a token in Settings, or create one at github.com/settings/tokens (no scopes required) — see the GitHub docs.
Tools are registered programmatically via the Python SDK. This list is read-only.