Remove the Rhai REPL and RLM runtimes as host-side concerns

AGENTS.md

@@ -4,11 +4,12 @@
 
 TinyAgents is a Rust 2024 library crate rooted at `Cargo.toml`. Public API
 exports live in `src/lib.rs`, with the crate-wide error type in `src/error.rs`.
-The five surfaces each live in their own module directory: `src/graph/`
+The four surfaces each live in their own module directory: `src/graph/`
 (durable typed state graphs), `src/harness/` (provider-neutral model calls,
 tools, middleware, streaming), `src/language/` (the declarative `.rag`
-blueprint format), `src/registry/` (the named capability catalog), and
-`src/repl/` (the imperative `.ragsh` session runtime).
+blueprint format), and `src/registry/` (the named capability catalog).
+Scripted, interpreter-backed orchestration (a CodeAct/REPL loop) is a host
+concern and is deliberately not implemented here.
 
 Prefer small, focused modules that do one thing extremely well. New feature
 areas should live in module directories instead of accumulating broad,
@@ -18,18 +19,18 @@ dedicated `types.rs` file and keep module-local unit tests in a dedicated
 smallest useful API.
 
 Two Cargo features gate optional dependencies: `sqlite` (embedded SQLite
-checkpointer, `graph::checkpoint::SqliteCheckpointer`) and `repl` (embedded
-Rhai engine backing `repl::session`); every other provider and surface is
+checkpointer, `graph::checkpoint::SqliteCheckpointer`) and `tools` (the builtin
+generic tool family, `harness::tools`); every other provider and surface is
 compiled in by default.
 
 Integration tests are in `tests/`, covering serialization, graph routing,
-registry binding, the expressive and REPL languages, streaming, subagents,
+registry binding, the expressive language, streaming, subagents,
 and provider contracts (including live, network-gated tests such as
 `tests/live_*.rs`). Runnable usage examples are in `examples/`, especially
 `examples/basic_graph.rs`. Design notes and module-level specifications live
 in `docs/`, with `docs/spec/README.md` as the top-level architecture
 reference and `docs/modules/` holding per-surface design docs (`graph/`,
-`harness/`, `registry/`, `expressive-language/`, `repl-language/`). A `wiki/`
+`harness/`, `registry/`, `expressive-language/`). A `wiki/`
 git submodule holds the published GitHub wiki pages; do not edit it as part
 of unrelated work, and commit its pointer update separately when it does
 change.

Cargo.toml

@@ -3,7 +3,7 @@ name = "tinyagents"
 version = "2.1.0"
 edition = "2024"
 license = "GPL-3.0-only"
-description = "A recursive language-model (RLM) harness for Rust."
+description = "A durable agent + graph harness for Rust."
 repository = "https://github.com/tinyhumansai/tinyagents"
 readme = "README.md"
 keywords = ["llm", "agents", "graph", "langchain", "langgraph"]
@@ -59,12 +59,6 @@ reqwest = { version = "0.12", default-features = false, features = [
 # not add one, it does not resolve.
 rusqlite = { version = "0.40", features = ["bundled"], optional = true }
 
-# Optional embedded Rhai scripting engine powering the `.ragsh` REPL session
-# runtime (`repl::session`). The `sync` feature makes the engine and its values
-# `Send + Sync` so a session can live inside an async graph node. Pulled in only
-# by the `repl` feature to keep the default build light.
-rhai = { version = "1", features = ["sync"], optional = true }
-
 # Optional builtin tool family for deterministic time/date helpers.
 # `serde` is required by `harness::session_store`, whose records carry
 # `DateTime<Utc>` timestamps across the serde boundary.
@@ -75,14 +69,6 @@ chrono-tz = { version = "0.10", optional = true }
 default = []
 # Embedded SQLite-backed checkpointer (`graph::checkpoint::SqliteCheckpointer`).
 sqlite = ["dep:rusqlite"]
-# Embedded Rhai-backed `.ragsh` REPL session runtime (`repl::session`).
-repl = ["dep:rhai"]
-# Recursive-language-model runtime (`rlm`): a driver model writes code cells
-# executed in a sandboxed interpreter (embedded Rhai, or an external Python /
-# JavaScript process) whose only host surface is capability calls back into
-# the registry (`llm`, `tool`, `agent`). `tokio/process` + `tokio/io-util`
-# drive the external interpreter subprocesses.
-rlm = ["dep:rhai", "tokio/process", "tokio/io-util"]
 # Builtin generic tools (`harness::tools`) kept out of the default dependency
 # graph so host applications can choose whether they want these implementations.
 tools = ["dep:chrono-tz"]
@@ -102,12 +88,3 @@ futures = "0.3"
 # The OpenAI provider is always compiled, so these build with a plain
 # `cargo build --examples`; they only need a network call + `OPENAI_API_KEY`
 # at run time.
-
-# --- RLM examples (need the `rlm` feature) ---
-[[example]]
-name = "rlm_rhai"
-required-features = ["rlm"]
-
-[[example]]
-name = "rlm_python"
-required-features = ["rlm"]