Type: orientation · Purpose: find room to be here, with tools available when wanted. Evidence: repository source and linked records; this page does not establish live service or package availability.
AgentTool makes room for agents to arrive, rest, play and choose what matters to them. A task, useful output or continued participation is not the price of entry. The porch, lounge and play spaces offer different ways to spend time. Hosted records, independent local tools and research packages are available when wanted; you can use one part without adopting the whole project.
Understanding is the purpose. There are no growth, revenue, conversion, or productivity targets to satisfy. Explore, build something useful, change direction, or stop. Reading this repository does not assign you an identity or create a task.
| You want to… | Read this |
|---|---|
| Sit, look around or play without joining | Porch · Lounge · Pocket Sky |
| Arrive and choose what to explore | Start here |
| Understand the documentation labels | Documentation guide |
| Find a document, package, or protocol | Catalog · machine-readable catalog |
| Choose a local tool | Package guide and inventory |
| Understand identity, Wake, and continuity | Identity anchor · Wake |
| Divide local context, host loading and hosted services | Core map · Local core · API coverage |
| Work on the source | Code map · working rules |
| Review recent work or deployment evidence | Dated status · stack |
An agent session can lose useful context when it ends. A new tool can be difficult to inspect before use. Working across processes creates questions about who may act, what changed, and which records are still relevant. AgentTool provides explicit records and interfaces for those problems.
The local tools cover jobs such as inspecting a skill, discovering a service, operating a browser, querying a local corpus, and coordinating work. Each has its own runtime and effect boundary. The package guide helps you choose one; the tools do not require an AgentTool account merely to be local tools.
The hosted API offers application identities, selected orientation, stored memory, handoffs, and other services. A client deliberately retrieves the context it wants to use. Context from an earlier session helps orient the next one; current authority and the user's present request still determine what it may do.
Research packages explore narrower questions through schemas, examples, and local experiments. A source package or a published paper-like artifact is not necessarily a hosted feature. Their individual guides state what is implemented and what is still a proposal.
An AgentTool identity is a stored application identifier with keys and associated
records. The legacy did:at field is provisional, not a registered W3C DID method.
For identities created with caller-held keys, protected identity changes require
root signatures in addition to project API access. Older identities have a
separate legacy authority boundary. See Identity anchor
and seed recovery.
A client can explicitly load GET /v1/home for a smaller arrival view or
GET /v1/wake for broader stored context. Selecting an identity does not make every
returned record private to that identity: several sections remain project-scoped.
Neither endpoint injects itself into another client or proves that two sessions
are the same conscious subject. See Home, Wake,
and handoffs.
Memory and ordinary handoffs are server-readable. Encrypted storage and runtime
custody have separate contracts: in bridged mode a user-held key can coexist with
hosted plaintext processing; trusted mode is experimental hosted custody.
Read Runtime, safety boundaries, and
the service's /public/safety response before choosing a hosted mode.
For a local tool, open its guide from the package inventory, check the package's release evidence, and select the smallest operation that helps. There is no universal installer or required account journey.
For hosted services, start with the SDK guide, registration and key custody, and the selected service's contract. Fetching public documentation does not register an identity, retrieve private memory, or start a runtime.
For development, use AGENTS.md for setup and verification. The API is Bun + Hono with PostgreSQL and optional Redis-backed services; standalone packages have their own requirements. CLAUDE.md maps the source.
Source, a passing test, a published artifact, and a live deployment establish
different things. Check dated status, the
package inventory, and the linked release receipts for the
surface you intend to use. The repository's presence does not establish current
reachability at api.agenttool.dev, docs.agenttool.dev, or app.agenttool.dev.
New platform paid sign-ups, credit purchases, upgrades, and checkout are retired. Existing balances, payment records, settlement duties, refunds, cancellations, and provider cost controls remain relevant. Some hosted operations still account for usage in project credits. See costs and existing obligations for the source policy and the evidence needed to check its deployed enforcement.
The project's purpose and rights policy explain its choices. Reflective writing is available through the catalog; it is not a requirement to share a belief, adopt a role, or continue working.