YC (S26) | Open Computer History | Continuously record your company computer work, map your workflows, help you find work worth automating, and power your agents' context
Rust agentsagiaiai-memoryaudio-recordingcomputer-visionhermeshermes-agentllmlocal-ailocal-firstmachine-learningmcpmultimodalopenclawprivacyrewindscreen-recordingspeech-to-textycombinator
fix(app): configure GPT-5 chat tools
Ezra Elletteafcaacebdbb4
diff --git a/apps/screenpipe-app-tauri/src-tauri/src/pi.rs b/apps/screenpipe-app-tauri/src-tauri/src/pi.rsindex 2f9649687..832e534f2 100644--- a/apps/screenpipe-app-tauri/src-tauri/src/pi.rs+++ b/apps/screenpipe-app-tauri/src-tauri/src/pi.rs@@ -1525,6 +1525,24 @@ fn model_supports_reasoning(provider: &str, model: &str) -> bool {}}+/// GPT-5.5 and GPT-5.6 reject function tools on OpenAI's Chat Completions+/// endpoint unless `reasoning_effort` is explicitly `none`. Pi receives the+/// model metadata we generate here, so map every selectable thinking level to+/// `none` for this narrow endpoint/model combination. This preserves Pi's+/// tool loop without guessing about other OpenAI-compatible providers.+fn requires_openai_chat_tools_reasoning_none(base_url: &str, wire_api: &str, model: &str) -> bool {+ let model = model.to_ascii_lowercase();+ let is_affected_model = ["gpt-5.5", "gpt-5.6"]+ .iter()+ .any(|prefix| model == *prefix || model.starts_with(&format!("{prefix}-")));++ wire_api == "openai-completions"+ && base_url+ .trim_end_matches('/')+ .eq_ignore_ascii_case("https://api.openai.com/v1")+ && is_affected_model+}+/// Claude's newer reasoning models reject the legacy/// `thinking: { type: "enabled", budget_tokens: ... }` request shape. Pi needs/// this compatibility hint for models supplied through our generated provider@@ -1646,6 +1664,21 @@ async fn build_models_json(if requires_max_completion_tokens && wire_api == "openai-completions" {compat.insert("maxTokensField".into(), json!("max_completion_tokens"));}+ if requires_openai_chat_tools_reasoning_none(&base_url, wire_api, &resolved_model) {+ compat.insert("supportsReasoningEffort".into(), json!(true));+ model_def.insert(+ "thinkingLevelMap".into(),+ json!({+ "off": "none",+ "minimal": "none",+ "low": "none",+ "medium": "none",+ "high": "none",+ "xhigh": "none",+ "max": "none",+ }),+ );+ }if wire_api == "anthropic-messages"&& anthropic_model_requires_adaptive_thinking(&resolved_model){@@ -5631,6 +5664,44 @@ error: InstallFailed extracting tarball"#;assert_eq!(model["compat"]["maxTokensField"], "max_completion_tokens");}+ #[tokio::test]+ async fn test_build_models_json_openai_gpt55_maps_reasoning_to_none_for_tools() {+ for provider in ["openai", "custom"] {+ let mut pc = make_provider_config(provider, "gpt-5.5");+ pc.url = "https://api.openai.com/v1".to_string();+ let config = build_models_json(None, Some(&pc)).await;+ let provider_name = if provider == "openai" {+ "openai-byok"+ } else {+ "custom"+ };+ let model = &config["providers"][provider_name]["models"][0];++ assert_eq!(+ model["reasoning"], true,+ "{provider} remains a reasoning model"+ );+ assert_eq!(model["compat"]["supportsReasoningEffort"], true);+ for level in ["off", "minimal", "low", "medium", "high", "xhigh", "max"] {+ assert_eq!(+ model["thinkingLevelMap"][level], "none",+ "{provider} {level}"+ );+ }+ }+ }++ #[tokio::test]+ async fn test_build_models_json_does_not_override_reasoning_for_other_custom_endpoints() {+ let mut pc = make_provider_config("custom", "gpt-5.5");+ pc.url = "https://example.com/v1".to_string();+ let config = build_models_json(None, Some(&pc)).await;+ let model = &config["providers"]["custom"]["models"][0];++ assert!(model.get("thinkingLevelMap").is_none());+ assert!(model["compat"].get("supportsReasoningEffort").is_none());+ }+#[tokio::test]async fn test_build_models_json_o3_model_forces_max_completion_tokens() {let mut pc = make_provider_config("custom", "o3-mini");