Skip to content

A self-improving RLM agent for coding workflows and long-running autonomous tasks.

Rust

use a non-reasoning model for session recaps (#284)

kt5c3b491dedd5
diff --git a/packages/ai/scripts/generate-models.ts b/packages/ai/scripts/generate-models.ts
index 4f6a3037a1..ad472bf44c 100644
--- a/packages/ai/scripts/generate-models.ts
+++ b/packages/ai/scripts/generate-models.ts
@@ -146,6 +146,7 @@ const PRIME_INFERENCE_MODEL_METADATA: Record<string, PrimeInferenceModelMetadata
"openai/gpt-5.5": { contextWindow: 1050000, maxTokens: 128000, vision: true },
"prime-intellect/intellect-3": { contextWindow: 131072, maxTokens: 131072 },
"qwen/qwen3-235b-a22b-thinking-2507": { contextWindow: 262144, maxTokens: 4096 },
+ "qwen/qwen3-30b-a3b-instruct-2507": { contextWindow: 262144, maxTokens: 65536 },
"qwen/qwen3-coder-next": { contextWindow: 262144, maxTokens: 65536 },
"qwen/qwen3-max": { contextWindow: 262144, maxTokens: 65536 },
"qwen/qwen3-vl-235b-a22b-thinking": { contextWindow: 262144, maxTokens: 32768, vision: true },
diff --git a/packages/ai/src/models.generated.ts b/packages/ai/src/models.generated.ts
index d562b339d1..1b3f1aeb95 100644
--- a/packages/ai/src/models.generated.ts
+++ b/packages/ai/src/models.generated.ts
@@ -13632,6 +13632,24 @@ export const MODELS = {
contextWindow: 131072,
maxTokens: 131072,
} satisfies Model<"openai-completions">,
+ "qwen/qwen3-30b-a3b-instruct-2507": {
+ id: "qwen/qwen3-30b-a3b-instruct-2507",
+ name: "QWEN3 30B A3B Instruct 2507",
+ api: "openai-completions",
+ provider: "prime-inference",
+ baseUrl: "https://api.pinference.ai/api/v1",
+ compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":true,"maxTokensField":"max_tokens","supportsStrictMode":false},
+ reasoning: false,
+ input: ["text"],
+ cost: {
+ input: 0.2,
+ output: 0.8,
+ cacheRead: 0,
+ cacheWrite: 0,
+ },
+ contextWindow: 262144,
+ maxTokens: 65536,
+ } satisfies Model<"openai-completions">,
"qwen/qwen3-coder-next": {
id: "qwen/qwen3-coder-next",
name: "QWEN3 Coder Next",
diff --git a/packages/coding-agent/src/modes/daemon/daemon-session-summarizer.ts b/packages/coding-agent/src/modes/daemon/daemon-session-summarizer.ts
index d3030f5f3c..0cbdf11e22 100644
--- a/packages/coding-agent/src/modes/daemon/daemon-session-summarizer.ts
+++ b/packages/coding-agent/src/modes/daemon/daemon-session-summarizer.ts
@@ -9,9 +9,8 @@ const SWEEP_INTERVAL_MS = 25_000;
// Collapse a tool-use loop's rapid turn_end bursts into one summarization.
const SETTLE_DEBOUNCE_MS = 2_000;
-// Small self-hosted open-weight model, off the proxied frontier models.
const SUMMARY_MODEL_PROVIDER = "prime-inference";
-const SUMMARY_MODEL_ID = "nvidia/nemotron-3-nano-30b-a3b";
+const SUMMARY_MODEL_ID = "qwen/qwen3-30b-a3b-instruct-2507";
const SUMMARY_CONTEXT_MESSAGES = 8;
const SUMMARY_MAX_CHARS_PER_MESSAGE = 600;