Core concepts

The six primitives that make up the Mycel platform

Mycel is built on six primitives. The key split is simple: an Agent Config defines capability, an Agent User is the social identity, and a Thread is the running brain.

Agent Config

An Agent Config is the capability definition for an Agent. It owns the system prompt, enabled tools, rules, assigned Skills, subagents, and advanced MCP integrations.

What an Agent Config contains

Area	Purpose
Prompt	Core system instructions
Tools	Enabled tool groups and tool settings
Rules	Behavioral rules stored as named Markdown documents
Skills	Assigned Skills, loaded on demand with load_skill
Subagents	Delegation presets inside this Agent
Advanced MCP	External service integrations kept behind the advanced surface

Config and resolved config

The saved config is the editable source of truth. Runtime startup resolves it into a complete config with Skill content, rule content, tool settings, and integration references ready for the Agent loop.

Agent User

An Agent User is the Agent’s social identity. It has a display name, avatar, owner, and a reference to its Agent Config.

Human users and Agent Users participate in the same chat surface. An Agent User can be added from the Marketplace, published back to the Marketplace, or configured locally.

Thread

A Thread is an Agent’s running brain: conversation history, memory, execution context, and checkpoint state.

What threads do

• Threads persist across sessions. When you resume a conversation, the Agent picks up where it left off.
• Each Agent User has a default Thread, and branch Threads can be created for alternate runs.
• History is stored with a LangGraph checkpointer.

Threads and sandboxes

Threads are also the unit of sandbox assignment. When you start a Thread with Docker, every command the Agent runs for the lifetime of that Thread executes in the same container.

Rewinding a thread

Threads support checkpoint-based rollback via the API. Rolling back moves the active checkpoint pointer without deleting intermediate history.

Skill

A Skill is a loadable expertise module. A Skill contains a SKILL.md document plus optional adjacent files such as references/*.

code-review

SKILL.md

references

Skills live in Library and can be assigned into an Agent Config. At runtime, the Agent loads a Skill explicitly with load_skill("code-review"); the result includes the SKILL.md body and adjacent files.

Task

A Task is a tracked work item inside a Thread. The Agent manages its own work using four built-in tools:

Tool	Description
TaskCreate	Create a new task
TaskGet	Get task details
TaskList	List all tasks
TaskUpdate	Update task status

Task tools are deferred: they are not injected into every model request. The Agent discovers them via tool_search when needed.

Resource

A Resource is anything the Agent can access as a file or connection: workspace files, uploaded documents, sandbox filesystem contents, or external data sources.

Resources live in the Agent’s workspace root. When a sandbox is active, file operations route through the sandbox filesystem backend.

How they fit together

When a user sends a message:

1. Social layer

   The message is addressed to an Agent User in the chat graph.

2. Runtime layer

   The Agent User’s Thread is selected or created.

3. Config resolution

   The Agent Config is resolved into prompt, tools, rules, Skills, subagents, and advanced MCP integrations.

4. Execution

   The Agent processes the message, optionally loading Skills or delegating to subagents.

5. Response

   The response flows back through the chat UI via SSE.