opencode-llmstack 0.9.2__py3-none-any.whl → 0.9.4__py3-none-any.whl

llmstack/__init__.py

@@ -16,5 +16,5 @@ organised by concern:
 
 from __future__ import annotations
 
-__version__ = "0.9.2"
+__version__ = "0.9.4"
 __all__ = ["__version__"]

llmstack/app.py

@@ -355,9 +355,10 @@ def classify(body: dict[str, Any]) -> tuple[str, str]:
         return AGENT_MODEL, f"ultra-trigger->agent ({ULTRA_MODEL} unavailable)"
 
     n_turns = sum(1 for m in (messages or []) if m.get("role") == "user")
+    _last_msgs = [{"role": "user", "content": last_user}] if last_user else None
     has_code_signal = (
-        _matches(CODE_BLOCK, messages, prompt)
-        or _matches(AGENT_SIGNALS, messages, prompt)
+        _matches(CODE_BLOCK, _last_msgs, prompt)
+        or _matches(AGENT_SIGNALS, _last_msgs, prompt)
     )
 
     est = _estimate_tokens(messages, prompt)

llmstack/backends/bedrock.py

@@ -400,7 +400,7 @@ def _tool_config(tools: list[dict[str, Any]] | None) -> dict[str, Any] | None:
     return {"tools": specs}
 
 
-def _inference_config(body: dict[str, Any]) -> dict[str, Any]:
+def _inference_config(body: dict[str, Any], max_output_tokens: int | None = None) -> dict[str, Any]:
     # We forward only what the Converse `inferenceConfig` schema accepts:
     # `temperature`, `topP`, `maxTokens`, `stopSequences`. Other sampler
     # knobs (`top_k`, `min_p`, `repetition_penalty`) have no Converse-
@@ -415,28 +415,34 @@ def _inference_config(body: dict[str, Any]) -> dict[str, Any]:
     # forward. Configure Bedrock tiers in models.ini accordingly: omit
     # the `sampler =` line for Opus 4.7+, and pick the one allowed knob
     # for Sonnet 4.5 / Haiku 4.5.
-    cfg: dict[str, Any] = {}
+    #
+    # `max_output_tokens` is the per-tier cap from models.ini
+    # (`aws_max_output_tokens`). When set, the client-requested value is
+    # silently clamped to this ceiling -- useful for models like
+    # Llama 3.1 405B whose Bedrock deployment rejects values above 4096.
+    icfg: dict[str, Any] = {}
     if "temperature" in body:
         try:
-            cfg["temperature"] = float(body["temperature"])
+            icfg["temperature"] = float(body["temperature"])
         except (TypeError, ValueError):
             pass
     if "top_p" in body:
         try:
-            cfg["topP"] = float(body["top_p"])
+            icfg["topP"] = float(body["top_p"])
         except (TypeError, ValueError):
             pass
     if "max_tokens" in body or "max_completion_tokens" in body:
         try:
-            cfg["maxTokens"] = int(body.get("max_tokens") or body.get("max_completion_tokens"))
+            requested = int(body.get("max_tokens") or body.get("max_completion_tokens"))
+            icfg["maxTokens"] = min(requested, max_output_tokens) if max_output_tokens else requested
         except (TypeError, ValueError):
             pass
     stop = body.get("stop")
     if isinstance(stop, str):
-        cfg["stopSequences"] = [stop]
+        icfg["stopSequences"] = [stop]
     elif isinstance(stop, list):
-        cfg["stopSequences"] = [s for s in stop if isinstance(s, str)]
-    return cfg
+        icfg["stopSequences"] = [s for s in stop if isinstance(s, str)]
+    return icfg
 
 
 def _build_converse_kwargs(tier: Tier, body: dict[str, Any], cfg: BedrockConfig) -> dict[str, Any]:
@@ -463,7 +469,7 @@ def _build_converse_kwargs(tier: Tier, body: dict[str, Any], cfg: BedrockConfig)
     if sys_blocks:
         converse_kwargs["system"] = sys_blocks
 
-    inference = _inference_config(body)
+    inference = _inference_config(body, max_output_tokens=cfg.max_output_tokens)
     if inference:
         converse_kwargs["inferenceConfig"] = inference
 

llmstack/models.ini

@@ -215,9 +215,10 @@ description  = Mistral-Small 3.2 24B Heretic - no-filter planning
 ; tier         = chat
 ; role         = plan-uncensored
 ; backend      = bedrock
-; aws_model_id = meta.llama3-1-405b-instruct-v1:0
-; aws_region   = us-west-2     ; Llama 405B has no EU deployment; keep on US
-; aws_profile  = bedrock-prod
+; aws_model_id         = meta.llama3-1-405b-instruct-v1:0
+; aws_region           = us-west-2     ; Llama 405B has no EU deployment; keep on US
+; aws_profile          = bedrock-prod
+; aws_max_output_tokens = 4096         ; Llama 3.1 405B Bedrock hard cap
 ; ctx_size     = 128000
 ; sampler      = temp=0.85, top_p=0.95   ; max exploration
 ; description  = Llama 3.1 405B on Bedrock - no-filter planning
@@ -229,10 +230,11 @@ description  = Mistral-Small 3.2 24B Heretic - no-filter planning
 ; tier             = chat
 ; role             = plan-uncensored
 ; backend          = bedrock
-; aws_model_id     = meta.llama3-1-405b-instruct-v1:0
-; aws_region       = us-west-2     ; Llama 405B has no EU deployment
-; aws_profile      = bedrock-prod
-; aws_endpoint_url = https://bedrock-runtime.us-west-2.vpce.amazonaws.com
+; aws_model_id         = meta.llama3-1-405b-instruct-v1:0
+; aws_region           = us-west-2     ; Llama 405B has no EU deployment
+; aws_profile          = bedrock-prod
+; aws_endpoint_url     = https://bedrock-runtime.us-west-2.vpce.amazonaws.com
+; aws_max_output_tokens = 4096         ; Llama 3.1 405B Bedrock hard cap
 ; ctx_size         = 128000
 ; sampler          = temp=0.85, top_p=0.95
 ; description      = Llama 3.1 405B on Bedrock (VPC) - no-filter planning

llmstack/tiers.py

@@ -139,6 +139,7 @@ class BedrockConfig:
     endpoint_url: str | None = None
     model_id_next: str | None = None
     region_next: str | None = None
+    max_output_tokens: int | None = None
 
     @property
     def has_next(self) -> bool:
@@ -298,6 +299,7 @@ def _build_bedrock(section) -> BedrockConfig:
         endpoint_url=_opt(section.get("aws_endpoint_url")),
         model_id_next=_opt(section.get("aws_model_id_next")),
         region_next=_opt(section.get("aws_region_next")),
+        max_output_tokens=_int(section.get("aws_max_output_tokens", "")) or None,
     )
 
 

{opencode_llmstack-0.9.2.data → opencode_llmstack-0.9.4.data}/data/CHANGELOG.md

@@ -4,6 +4,21 @@ All notable changes to `opencode-llmstack` are documented here.
 
 ---
 
+## [0.9.4] — 2026-05-11
+
+### Fixed
+- `classify()` now scopes `has_code_signal` to the **last user message only**
+  (was scanning the full conversation history). Previously, any prior coding
+  exchange in the session (code blocks, agent verbs) would permanently block
+  plan routing for the rest of the conversation — e.g. "explain why these
+  changes are important?" after a refactor request would never reach `plan`.
+- Added regression test:
+  `test_plan_signal_after_prior_coding_exchange_routes_to_plan`.
+- `__version__` corrected from `"0.9.2"` to `"0.9.4"` (was skewed vs
+  `pyproject.toml` since 0.9.3).
+
+---
+
 ## [0.9.2] — 2026-05-11
 
 ### Fixed

{opencode_llmstack-0.9.2.dist-info → opencode_llmstack-0.9.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: opencode-llmstack
-Version: 0.9.2
+Version: 0.9.4
 Summary: Multi-tier local LLM stack: llama-swap + FastAPI auto-router + opencode wiring.
 Author: llmstack
 License: MIT License

{opencode_llmstack-0.9.2.dist-info → opencode_llmstack-0.9.4.dist-info}/RECORD

@@ -1,16 +1,16 @@
 llmstack/AGENTS.md,sha256=4DVUkqJ1-EP-cDNRCpznzghOOX6dAMbVWdcwyfFCALw,528
-llmstack/__init__.py,sha256=Qs9d58V8cJWmJvu4QLvO7_panKa8UkGRXAurHYgKDyU,855
+llmstack/__init__.py,sha256=C56mGWt0YsaDyU0C7R_PpsIPw67G0jm4nRAqGKLVh5s,855
 llmstack/__main__.py,sha256=wXHd5-BmCCHUfNEmy2rbilBSyVhi4KD1dSIO_4NlxuE,199
 llmstack/_platform.py,sha256=eDY3T9krkaBigG5xXxqzIbH3MhdZqX3BWe7bozOsAso,13099
-llmstack/app.py,sha256=3Qt_bveLS13rPEucyX0P6QDf9o9O68amnJIvwMqoTQQ,28469
+llmstack/app.py,sha256=FpxEWpfvnRdRJg3IQdCOSdSuqPQG6iYKpBNGcfUR9wA,28554
 llmstack/check_models.py,sha256=WvTS2Td4acp-Q0-yWXUgXAgAgFOmpxiaeSDuAoivirw,4559
 llmstack/cli.py,sha256=Om70PzHrmU81y2Mw1sB6eeUs1fRHP0PnsCEVNC0UNvI,11341
-llmstack/models.ini,sha256=USyEOekc3tWw-m375K-zDc3EyaQjrJGEnG_Hf5ZxdWw,20370
+llmstack/models.ini,sha256=U38Z6wfGCpmgQTCNbnp1zu80rUrhNGWFTvI2nhVx1Mo,20556
 llmstack/paths.py,sha256=A8q4-tpwIt5UMGG5ZDESKSuViMGLbPIAL1VoONopJqU,11512
 llmstack/shell_env.py,sha256=MJSW0PP15q-fsppIZ98WZ7XoqYMZmDy4k8N0gzEA6wU,39362
-llmstack/tiers.py,sha256=et738dWftsc74ZElZ3Vt9eEF_SzgJCDuH9kBhzH-scI,14697
+llmstack/tiers.py,sha256=yl5xEhECe-GHiVXBRvlNoFtH_9y4uNSASpfHlZ4Ja74,14820
 llmstack/backends/__init__.py,sha256=-85sQz0R94OdbM2bUHGyyA5WaMnI9bHywPOaELeQHX0,777
-llmstack/backends/bedrock.py,sha256=eVYuAWyX7NeRi0X55H7kVBTJAF7hmutyTS2lrCzBZeY,30179
+llmstack/backends/bedrock.py,sha256=Nb9sV45aH0RQHie_AkQwcpX5pkio5EAqnsphZM5P_nQ,30638
 llmstack/commands/__init__.py,sha256=eVO-YUxh1fSfdq72KggC-NrTYMtN6zIykgjyRgOCAt4,406
 llmstack/commands/_helpers.py,sha256=UKADaNXrnuoDi_JG0W2Tph7rWFB0cXvQh8YknZBw56I,2660
 llmstack/commands/activate.py,sha256=zCdEmyVv5qZUdhfez6hZ5Y46N_yjPwfKbPTwCJXnA3o,3663
@@ -30,12 +30,12 @@ llmstack/download/ggufs.py,sha256=2hCr-svUiPIV2I3ruwTbXo6lPn9m-VBOqa3DFbvdIcA,54
 llmstack/generators/__init__.py,sha256=LfbcReuyYBCdVuT9J5RKo7-f8n585YBU3Hus6DsxqTs,1189
 llmstack/generators/llama_swap.py,sha256=KdYH9N6TJECotZvyxvAjaa3kRyzn4YOi2T6D2UdyVKw,14785
 llmstack/generators/opencode.py,sha256=s_FrLXUBnLzRGQovl1PcAEs7V_P52wT1vnvvxMcKfs4,11203
-opencode_llmstack-0.9.2.data/data/CHANGELOG.md,sha256=2Ok5sn4aA_N5UMaUJ4jRbuTeWC1pt7gdFgYnxj2JdKU,5217
-opencode_llmstack-0.9.2.data/data/LICENSE,sha256=6G-Otw6BHIM1WJSBlJ04P1rDVCqbDEzKpdOlSr5CqIY,1078
-opencode_llmstack-0.9.2.data/data/UPGRADING.md,sha256=0XSNZ9trCviFLH5EL3Jz02fO2_8AfqB8_9aX0-o1bik,24927
-opencode_llmstack-0.9.2.dist-info/licenses/LICENSE,sha256=6G-Otw6BHIM1WJSBlJ04P1rDVCqbDEzKpdOlSr5CqIY,1078
-opencode_llmstack-0.9.2.dist-info/METADATA,sha256=4HqjCNQ363c6DSkSuOZWSYxDvIujqhEc-DmRFNPhfNI,36323
-opencode_llmstack-0.9.2.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
-opencode_llmstack-0.9.2.dist-info/entry_points.txt,sha256=soomjpqvl4KzFScgpQbu96vgcLriOtkB9MbiSC0rvZ8,47
-opencode_llmstack-0.9.2.dist-info/top_level.txt,sha256=tMv9sDWp8RW_DNNY8cuM4Uy4sND-KwTLcsScl5gdcEQ,9
-opencode_llmstack-0.9.2.dist-info/RECORD,,
+opencode_llmstack-0.9.4.data/data/CHANGELOG.md,sha256=58feU0rA9bBYvecDoFaLcwwgezLPkD3MSt0vRUVjdF8,5837
+opencode_llmstack-0.9.4.data/data/LICENSE,sha256=6G-Otw6BHIM1WJSBlJ04P1rDVCqbDEzKpdOlSr5CqIY,1078
+opencode_llmstack-0.9.4.data/data/UPGRADING.md,sha256=0XSNZ9trCviFLH5EL3Jz02fO2_8AfqB8_9aX0-o1bik,24927
+opencode_llmstack-0.9.4.dist-info/licenses/LICENSE,sha256=6G-Otw6BHIM1WJSBlJ04P1rDVCqbDEzKpdOlSr5CqIY,1078
+opencode_llmstack-0.9.4.dist-info/METADATA,sha256=hHxDa4-Iqb7saWESfg82n4DEhFpM0J9uY9hAosnuJAQ,36323
+opencode_llmstack-0.9.4.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
+opencode_llmstack-0.9.4.dist-info/entry_points.txt,sha256=soomjpqvl4KzFScgpQbu96vgcLriOtkB9MbiSC0rvZ8,47
+opencode_llmstack-0.9.4.dist-info/top_level.txt,sha256=tMv9sDWp8RW_DNNY8cuM4Uy4sND-KwTLcsScl5gdcEQ,9
+opencode_llmstack-0.9.4.dist-info/RECORD,,