Glossary
Key terms and concepts in Astonish
Key terms and concepts used throughout Astonish, listed alphabetically.
The AI entity you interact with in chat. Uses tools and memory to accomplish tasks.
Channel
Section titled “Channel”A communication adapter connecting Astonish to external messaging platforms (Telegram, Email).
Compaction
Section titled “Compaction”Automatic summarization of older conversation messages when the context window gets full.
Credential
Section titled “Credential”An encrypted secret stored in the credential store. Used for API authentication, OAuth, and service access.
Daemon
Section titled “Daemon”The Astonish background service that runs Studio, channels, scheduler, and fleet sessions.
Delegation
Section titled “Delegation”When the agent spawns sub-agents to handle parts of a task in parallel.
Distillation
Section titled “Distillation”Converting an interactive chat session into a reusable YAML flow.
A connection between nodes in a flow. Can be unconditional or conditional.
Embedding
Section titled “Embedding”A vector representation of text used for semantic search in the memory system.
Multi-agent team system. A group of specialized AI agents that collaborate on complex tasks.
A structured automation workflow defined in YAML as a directed graph of nodes and edges.
Knowledge
Section titled “Knowledge”Information stored in the memory system. Retrieved automatically or via memory_search.
Master Key
Section titled “Master Key”An optional password protecting credential store values from being viewed.
Model Context Protocol — an open standard for connecting AI agents to external tool servers.
Memory
Section titled “Memory”Persistent vector store where Astonish saves and retrieves knowledge across sessions.
A unit of work in a flow (input, llm, tool, output, or update_state).
A configured fleet template instance bound to a specific project, channel, and credentials.
Provider
Section titled “Provider”An AI model service (OpenAI, Anthropic, Ollama, etc.) that Astonish connects to for inference.
Redaction
Section titled “Redaction”The 5-layer security system that prevents secrets from appearing in logs or LLM context.
Scheduler
Section titled “Scheduler”The cron-based job execution system that runs tasks on a schedule.
Session
Section titled “Session”A persisted conversation with its full history, stored as a JSONL file.
A pre-built instruction guide that teaches the agent how to use a specific CLI tool or workflow.
The shared key-value store that passes data between nodes in a flow.
Sub-agent
Section titled “Sub-agent”A child agent spawned by delegation, with its own session and filtered tool access.
Sub-session
Section titled “Sub-session”A session created by a sub-agent, linked to its parent session.
A Git repository that provides flows and MCP server configurations as extensions.
Template
Section titled “Template”A fleet team definition specifying agent roles, models, and capabilities.
Thread
Section titled “Thread”A pairwise conversation between two agents in a fleet session.
A capability the agent can invoke (file operations, shell commands, web requests, etc.).
Toolset
Section titled “Toolset”A group of related tools (e.g., browser tools, email tools).
Vector Store
Section titled “Vector Store”The chromem-go database that stores embeddings for semantic memory search.