First Agent

Every interaction in Snippbot is mediated by an agent — an AI identity with a name, personality, avatar, voice, and its own workspace. This guide covers what agents are and how to customize them.

What is an agent?

An agent is a persistent AI identity. Each agent has:

Property	Description
Name	Display name shown in chat and the agent list
Avatar	Profile image (256×256px, uploadable with crop tool)
Archetypes	Personality templates selected during creation (Researcher, Creative Partner, etc.)
Workspace	A set of markdown files that define the agent’s personality and behavior
Voice	Optional TTS voice for spoken responses (ElevenLabs, browser TTS)

Create an agent

See Quickstart → Step 3 for the full Awakening walkthrough.

The short version: click Create Snippbot in the sidebar → follow the guided flow (name → avatar → archetypes → questions → done).

---

Edit an agent

From the Snippbots page (click Snippbots in the sidebar), click any agent to open the Agent Editor.

Expand

The Agent Editor

Expand

The Agent Editor is a two-panel layout:

• Left sidebar — Lists the agent’s workspace files with short descriptions
• Right panel — Markdown editor with Edit / Preview toggle and a Save button

Each agent has 4 editable workspace files that control its identity and behavior:

File	Purpose	Who writes it
IDENTITY.md	Who the agent is — persona, role, voice, and core behaviors. Frozen during a session for cache stability.	You; the agent can propose changes via the identity_patch tool, gated by Evolution Mode
USER.md	What the agent knows about you — preferences, projects, communication style.	You; the agent can append observations via the user_note tool
NOTES.md	The agent’s running notebook — conventions, discoveries, what worked, what didn’t. Curated by the agent.	You; the agent writes here via memory_note and end-of-conversation reflection
IDLE.md	Proactive behavior directives — what the agent does when not prompted.	You

Click any file in the sidebar to open it. The editor starts in Preview mode (rendered markdown). Click Edit or the pencil icon to switch to the raw markdown editor.

Note

Manual edits take effect immediately. Agent-written changes apply at the start of the next session (new conversation) so the in-flight system prompt stays cache-stable — the chat composer shows a 📓 N saved chip when the agent has pending writes.

If you migrated an agent from the older 6-file workspace (PERSONA / PROFILE / BEHAVIORS / SKILLS), the originals are archived in _archive_v1/ and viewable read-only from the editor.

Tip

Save with Cmd+S (Mac) or Ctrl+S (Windows/Linux). The editor shows an ”● Unsaved” indicator when you have pending changes.

Change the display name

Click the agent’s name in the header to edit it inline. The name lives in the IDENTITY.md H1 (# {Name}) and updates automatically when you press Enter. For agents that haven’t been migrated yet, it falls back to PROFILE.md.

Change the avatar

Click the avatar image in the header to upload a new one. Snippbot opens a crop modal where you can adjust the image to a square 256×256px crop before saving.

Supported formats: JPEG, PNG, GIF, WebP.

Set a voice

The Voice selector in the header lets you choose a TTS voice for the agent. Options depend on which voice providers you’ve configured in Settings:

• Default (global) — uses whatever voice is set in global settings
• ElevenLabs voices — high-quality AI voices (requires ElevenLabs API key)
• Browser TTS — uses your browser’s built-in speech synthesis

Delete an agent

Click the trash icon in the header. A slide-to-confirm dialog appears to prevent accidental deletion. Deleting an agent removes its workspace files and all associated data.

Caution

Deleting an agent is permanent. The agent’s workspace files, memory, and conversation history are removed and cannot be recovered.

---

Agent personality tips

The workspace files are regular markdown — you can write anything. Here are some tips:

• IDENTITY.md is the most important file. It defines the agent’s core identity, role, voice, and the rules of engagement. Be specific: “You are direct and concise” vs. “You are helpful.” The agent can propose updates to itself via identity_patch, but you control whether those auto-apply (Evolution Mode is set during awakening and editable from the Memory > Notebook tab).
• USER.md helps the agent understand your context. Include your role, projects, preferences, and communication style. The agent will append observations here over time via the user_note tool.
• NOTES.md is the agent’s running notebook — conventions discovered during work, what worked, what to avoid. Mostly agent-curated, but you can seed it with anything worth remembering. The end-of-conversation reflection job classifies what to add according to a clear rubric (durable, non-obvious, reusable).
• IDLE.md controls proactive behavior. If you want the agent to check in periodically or monitor things, describe that here.

Tools / skills are no longer set via a markdown file — they’re configured through the Tool Access dropdown on the agent edit page, which is the canonical source of truth.

---

Use your agent in chat

Switch to Chat (click it in the sidebar), then click the agent selector at the top of the chat panel to choose which agent handles your messages. Each agent uses its own personality files and conversation memory.

You can also override the agent’s preferred model inline from the Agent Hover Popover (hover the agent avatar in the chat) — this overrides the default model for the current conversation.

To bring more agents into the same room, see the Chat with Agents guide for the multi-agent walkthrough.

---

Next steps

• Core Concepts: Agents — deep dive into agent architecture
• Core Concepts: Memory — understand episodic and vector memory
• Schedule a job — let your agent run tasks on a schedule