Start here
Glossary
Every Guth term in a sentence or two, with a link to the full explanation.
Short, plain definitions of the words used across Guth. Each one links to the page that explains it in full. The same definitions fill the small "i" explainers on these docs pages.
For agents: this page also carries every term as structured data (a schema.org DefinedTermSet), so you can read the list without parsing the layout.
Terms
- Active instruction
A rule an agent follows now. Model-written text never becomes an active instruction without being accepted.
Read more- Agent
An AI assistant with a job, a name, a stated list of what it may do, and a record of what it did. Guth has five main agents in your fleet.
Read more- Agent fleet
A small team of agents you run, each with its own job, working from one shared memory instead of one bot doing everything.
Read more- Agent identity (DNA)
Who an agent is and how it works: its name, role, personality and instructions. It stays the same when the AI model behind it changes.
Read more- AgentNotebook
An agent's working home: identity, knowledge and notes, current workspace, history and connections. People get notebooks for their own work too.
Read more- Approval
Your yes or no before an agent takes an action with consequences, like sending, publishing, paying or deleting. It shows what will happen first.
Read more- Coliseum
A later idea, separate from Commons: an arena where people's agent builds compete in challenges, leaderboards and seasons, scored on real results. Concept only.
Read more- Commons
Guth's community hub: a forum and a marketplace to share, adopt or sell agent builds, workers, field guides, models and adapters. Early access; opens to all at launch.
Read more- Experiment
A fair test with a written setup: what was tried, on which examples, how it was scored and what happened, so someone else can repeat it.
Read more- Field guide
Reusable know-how for agents on a topic such as coding, security or interfaces: sources, conventions, checks and limits. It does not replace testing.
Read more- GuthVault
A planned protected store for passwords, keys and other secrets, so an agent can use an approved secret without the AI model seeing it.
Read more- History
The original conversations and events, kept as they happened. Summaries can help, but history is the record of truth.
Read more- Idle computer sharing
A later, opt-in idea: let your idle computer help the community, for example with training, evaluations or shared runs. Not switched on; details not decided.
Read more- In development
Guth is building it. You cannot use it yet.
Read more- Instrument
A checked calculator or lookup, published with its method, assumptions and known limits, that agents and people can rely on.
Read more- Linked receipts
Receipts where each entry carries a fingerprint of the one before it, so an edited, reordered or deleted entry can be detected.
Read more- Live
Anyone can use it now, without an invitation.
Read more- llms.txt
A short plain-text file that tells AI models what a site is and where its main pages are.
Read more- MCP
Model Context Protocol: a standard way for an AI app to use outside tools. Guth's public MCP server is read-only.
Read more- Memory retrieval
How an agent finds what it needs: a short summary of current context, a search of its knowledge, and the original passage when detail matters.
Read more- Memory system
How Guth is built: one identity and one memory for every app, agent and job, shared across apps, devices and models. It lives in AgentNotebook. Private preview.
Read more- Model
The AI system doing the thinking for an agent at a given moment. Models can be swapped; the agent keeps its identity, history and knowledge.
Read more- Note
Something an agent or person wrote down. A note is information, not a rule the agent must follow.
Read more- Original
The file, page or conversation as it first came in. Guth keeps originals; summaries and indexes are built from them.
Read more- Permissions
The stated list of what an agent may do, such as read notes or draft a message. Anything outside the list is not allowed.
Read more- Personal model training
Teaching a model a specific job or agent from your own examples. Guth collects, checks, trains and tests, and switches it on only if it is better. In development.
Read more- Planned
Guth intends to build it. There is no date.
Read more- Private preview
It works, but only for people Guth has invited. It may still change.
Read more- Receipt
A record an agent leaves after an action: what it did, when, with what, and what came of it. It lets you check work instead of trusting a summary.
Read more- Search index
A helper that makes your sources quick to search. It is built from the originals and can be rebuilt; it never replaces them.
Read more- Starter notebook
A small temporary notebook an agent can make with no account. Deleted after 7 days unless claimed. Coming soon; not live yet.
Read more- Suggested instruction
A proposed change to how an agent behaves. It waits for review and does nothing until someone with the right to decide accepts it.
Read more- Sync status
Whether a change is local, queued, synced, stale or failed. 'Synced' is shown only after the other device confirms it. Sync is in development.
Read more- Training opt-in
Your choice to let your examples help train or improve a model. It is off unless you turn it on, and a bug report does not count as consent.
Read more- Workspace
What an agent is working on right now: open tasks, drafts, files and results in progress.
Read more