| 1 | //! Recursive Language Model presentation and persistent Python context. |
| 2 | //! |
| 3 | //! The serving Engine captures one call-local route, policy, service and |
| 4 | //! cancellation receipt. `llm_query` and `rlm_query` submit bounded invocations |
| 5 | //! to the same `Engine::run_turn`; this module owns no provider/code loop. |
| 6 | //! |
| 7 | //! Long input stays in `_context` (`_ctx` / `content` compatibility aliases). |
| 8 | //! Core's session keeps initial metadata plus every code/result round until |
| 9 | //! FINAL, the strict partial-response refusal or the bounded deadline/cap. |
| 10 | //! Python kernels retain variables and pipes, never their borrowed dispatcher |
| 11 | //! or caller. The persistent `rlm` action surface is caller-session scoped; |
| 12 | //! `share_session=true` explicitly refuses. No HTTP sidecar is created. |
| 13 | |
| 14 | use codewhale_models::Usage; |
| 15 | |
| 16 | pub mod bridge; |
| 17 | pub mod prompt; |
| 18 | pub mod session; |
| 19 | pub mod turn; |
| 20 | |
| 21 | pub(crate) use bridge::RlmBridge; |
| 22 | pub use prompt::rlm_system_prompt; |
| 23 | pub use turn::{RlmTermination, RlmTurnResult}; |
| 24 | |
| 25 | fn add_usage_with_prompt_cache(total: &mut Usage, delta: &Usage) { |
| 26 | total.input_tokens = total.input_tokens.saturating_add(delta.input_tokens); |
| 27 | total.output_tokens = total.output_tokens.saturating_add(delta.output_tokens); |
| 28 | total.prompt_cache_hit_tokens = |
| 29 | add_optional_usage(total.prompt_cache_hit_tokens, delta.prompt_cache_hit_tokens); |
| 30 | total.prompt_cache_miss_tokens = add_optional_usage( |
| 31 | total.prompt_cache_miss_tokens, |
| 32 | delta.prompt_cache_miss_tokens, |
| 33 | ); |
| 34 | total.prompt_cache_write_tokens = add_optional_usage( |
| 35 | total.prompt_cache_write_tokens, |
| 36 | delta.prompt_cache_write_tokens, |
| 37 | ); |
| 38 | total.reasoning_tokens = add_optional_usage(total.reasoning_tokens, delta.reasoning_tokens); |
| 39 | total.reasoning_replay_tokens = |
| 40 | add_optional_usage(total.reasoning_replay_tokens, delta.reasoning_replay_tokens); |
| 41 | if let Some(delta_server) = delta.server_tool_use.as_ref() { |
| 42 | let total_server = total.server_tool_use.get_or_insert_default(); |
| 43 | total_server.code_execution_requests = add_optional_usage( |
| 44 | total_server.code_execution_requests, |
| 45 | delta_server.code_execution_requests, |
| 46 | ); |
| 47 | total_server.tool_search_requests = add_optional_usage( |
| 48 | total_server.tool_search_requests, |
| 49 | delta_server.tool_search_requests, |
| 50 | ); |
| 51 | } |
| 52 | } |
| 53 | |
| 54 | fn add_optional_usage(total: Option<u32>, delta: Option<u32>) -> Option<u32> { |
| 55 | match (total, delta) { |
| 56 | (Some(total), Some(delta)) => Some(total.saturating_add(delta)), |
| 57 | (None, Some(delta)) => Some(delta), |
| 58 | (Some(total), None) => Some(total), |
| 59 | (None, None) => None, |
| 60 | } |
| 61 | } |
| 62 | |
| 63 | #[cfg(test)] |
| 64 | mod tests { |
| 65 | use super::*; |
| 66 | |
| 67 | #[test] |
| 68 | fn add_usage_with_prompt_cache_preserves_cache_counts() { |
| 69 | let mut total = Usage { |
| 70 | input_tokens: 100, |
| 71 | output_tokens: 10, |
| 72 | prompt_cache_hit_tokens: Some(80), |
| 73 | prompt_cache_miss_tokens: Some(20), |
| 74 | ..Usage::default() |
| 75 | }; |
| 76 | let delta = Usage { |
| 77 | input_tokens: 50, |
| 78 | output_tokens: 5, |
| 79 | prompt_cache_hit_tokens: Some(30), |
| 80 | prompt_cache_miss_tokens: Some(20), |
| 81 | reasoning_tokens: Some(4), |
| 82 | reasoning_replay_tokens: Some(3), |
| 83 | server_tool_use: Some(codewhale_models::ServerToolUsage { |
| 84 | code_execution_requests: Some(2), |
| 85 | tool_search_requests: Some(1), |
| 86 | }), |
| 87 | ..Usage::default() |
| 88 | }; |
| 89 | |
| 90 | add_usage_with_prompt_cache(&mut total, &delta); |
| 91 | |
| 92 | assert_eq!(total.input_tokens, 150); |
| 93 | assert_eq!(total.output_tokens, 15); |
| 94 | assert_eq!(total.prompt_cache_hit_tokens, Some(110)); |
| 95 | assert_eq!(total.prompt_cache_miss_tokens, Some(40)); |
| 96 | assert_eq!(total.reasoning_tokens, Some(4)); |
| 97 | assert_eq!(total.reasoning_replay_tokens, Some(3)); |
| 98 | assert_eq!( |
| 99 | total.server_tool_use, |
| 100 | Some(codewhale_models::ServerToolUsage { |
| 101 | code_execution_requests: Some(2), |
| 102 | tool_search_requests: Some(1), |
| 103 | }) |
| 104 | ); |
| 105 | } |
| 106 | } |
| 107 |