# `GenAgent.Backends.Anthropic`
[🔗](https://github.com/genagent/gen_agent/blob/gen_agent_anthropic-v0.2.3/integrations/anthropic/lib/gen_agent/backends/anthropic.ex#L1)

`GenAgent.Backend` implementation backed by the Anthropic Messages API.

Unlike the CLI-backed backends (`GenAgent.Backends.Claude`,
`GenAgent.Backends.Codex`), this backend talks directly to an HTTP
API. The **conversation history lives in the session struct**,
because the API is stateless -- every request carries the full
messages array.

This backend exists primarily as a **validation spike** for the
`GenAgent.Backend` contract. The CLI backends share a lot of DNA
with each other (subprocess + NDJSON + server-side session ids); an
HTTP backend inverts most of that. Getting both to fit the same
contract without changes is the real test of whether the
abstraction is right.

## How the two halves of a turn land in the session

1. `prompt/2` appends the user's message to `session.messages`
   **before** making the API call, and returns the updated session
   along with the event list. The state machine stores that updated
   session.
2. When the state machine delivers the terminal `:result` event,
   it calls `update_session/2` with the event's data, and this
   backend appends the assistant's message to `session.messages`.
3. The next `prompt/2` sees both messages in the session and sends
   the full history to the API.

This uses both sides of the `GenAgent.Backend` contract in a way
the CLI backends don't: CLI backends leave `prompt/2`'s returned
session unchanged and do all updates in `update_session/2`.

## Options

  * `:api_key` -- Anthropic API key. Defaults to `System.get_env("ANTHROPIC_API_KEY")`.
  * `:model` -- model name. Defaults to `"claude-sonnet-4-5"`.
  * `:max_tokens` -- max tokens per turn. Defaults to `1024`.
  * `:system` -- system prompt (string).
  * `:receive_timeout` -- HTTP receive timeout in milliseconds.
    Defaults to `60_000`. Long-context turns (big messages array,
    slow models) can blow through Req's 15s default, so the backend
    picks a safer default. Set higher for large debates or longer
    generations.
  * `:connect_timeout` -- HTTP connect timeout in milliseconds.
    Defaults to Req's default when unset.
  * `:http_fn` -- a 1-arity function `(request_map) -> {:ok, response_map} | {:error, term}`
    that replaces the default `Req`-backed HTTP call. Intended for tests.

# `message`

```elixir
@type message() :: %{role: String.t(), content: String.t()}
```

# `t`

```elixir
@type t() :: %GenAgent.Backends.Anthropic{
  api_key: String.t() | nil,
  client_session_id: String.t(),
  connect_timeout: timeout() | nil,
  http_fn: (map() -> {:ok, map()} | {:error, term()}),
  max_tokens: pos_integer(),
  messages: [message()],
  model: String.t(),
  receive_timeout: timeout(),
  system: String.t() | nil
}
```

---

*Consult [api-reference.md](api-reference.md) for complete listing*