thal 0.0.1

Reactive semantic runtime — molecules, reactions, and effect actors for building LLM-backed applications as dataflow programs.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49

// Three-reaction LLM chat loop, plus an Init reaction that seeds the loop.
//
// The cycle TerminalPrompt -> LlmCall -> TerminalWrite -> TerminalPrompt is
// allowed by the verifier because every step crosses a builtin Effect actor
// that paces the loop on external I/O (user input, LLM HTTP, stdout flush).
//
// Default `provider: "mock"` echoes the prompt back. To use a real LLM:
//
//   1. Run whatever auth tool produces a token file. For Codex, use Codex
//      CLI's login. For a manual paste: `thal login codex`.
//   2. Edit Init to emit an `LlmProvider` molecule alongside the
//      `TerminalPrompt`, pointing at the token file and base URL.
//   3. Switch `OnUserInput`'s provider field to "codex".
//   4. Run `cargo run -- examples/chat.thal`.
//
// If you skip step 2 and use a non-mock provider, the runtime will prompt
// you for a token file path on first use and register the provider for you.

reaction OnUserInput {
    when:  TerminalPrompt(p)
    where: p.status == "Done"
    emit:  LlmCall { provider: "codex", model: "gpt-5.3-codex", prompt: p.answer }
}

reaction StartThinking {
    when:  LlmCall(c)
    where: c.status == "Pending"
    emit:  Spinner { label: "thinking…", running: true }
}

reaction OnLlmResponse {
    when:  LlmCall(c)
    where: c.status == "Done"
    emit:
        Spinner { label: "", running: false },
        TerminalWrite { content: c.text, markdown: true }
}

reaction OnPrintDone {
    when:  TerminalWrite(w)
    where: w.status == "Done"
    emit:  TerminalPrompt { question: "❯ " }
}

reaction Init {
    when: Boot(b)
    emit:
        TerminalPrompt { question: "❯ " }
}