--- name: lm-studio-subagents-822583e78e65 description: >- Offload tasks to local LLMs via LM Studio. Use when a user asks to run local models with LM Studio, save API costs by using local LLMs, create subagents with local models, offload summarization or classification to a local model, or use LM Studio's API for batch processing. Covers local model inference, task delegation, and cost optimization. license: Apache-2.0 compatibility: "Requires LM Studio installed and running locally; the native /api/v1/chat endpoint requires LM Studio 0.4.0+" metadata: author: terminal-skills version: "1.2.0" category: data-ai repository: https://github.com/lmstudio-ai/lmstudio-js tags: "lm-studio, local-llm, subagent, inference, cost-saving" use-cases: "Offload repetitive LLM tasks to free local models, Run summarization and classification without API costs, Create multi-agent workflows mixing cloud and local models" agents: "claude-code, openai-codex, gemini-cli, cursor" --- # LM Studio Subagents ## Overview Offload LLM tasks to local models running in LM Studio to save API costs and maintain privacy. LM Studio provides a native REST API and OpenAI-compatible endpoints. Prefer the native `/api/v1/chat` API for long-form work because it exposes output-token, reasoning, context-length, streaming, and stateful-continuation controls. Use local models for high-volume, lower-complexity tasks like summarization, extraction, classification, and reformatting while reserving cloud APIs for complex reasoning. ## Instructions When a user wants to use local models via LM Studio, determine the task: ### Task A: Set up LM Studio as a local API server 1. Download and install LM Studio from `https://lmstudio.ai/` 2. Download a model: in the app use Model Search, or from a terminal run `lms get` (no argument lists curated recommendations; `lms get llama-3.1-8b@q4_k_m` picks a quantization). Instruct models in the 4B-8B range (Qwen, Llama, Mistral families) are a good start; check the memory estimate first with `lms load --estimate-only qwen2.5-7b-instruct`. 3. Start the local server: toggle "Start server" in the Developer tab, or run `lms server start` (add `--port 1234` to pin the port; the CLI otherwise reuses the last port, and 1234 is the usual default). It listens on `127.0.0.1` only unless you pass `--bind`; `--cors` and non-local binds need authentication switched on. 4. Load a model with a stable API name and context size, then verify: ```bash lms load qwen2.5-7b-instruct --identifier local-worker --context-length 16384 curl http://localhost:1234/v1/models ``` The `id` values returned are what you pass as `model`. Export it: `export LM_MODEL=local-worker`. 5. Authentication is off by default ("does not require authentication"). To enable it, go to Developer, Server Settings, switch authentication on, then Manage Tokens, Create Token, and copy the token once. Export it as `LM_API_TOKEN`; send it as `Authorization: Bearer $LM_API_TOKEN`. With auth off, the header is harmless. OpenAI-compatible clients still need some non-empty key string. ### Task B: Call LM Studio from Python (native v1 API, preferred) The native endpoint exposes output-length, reasoning, and stateful-continuation controls the OpenAI-compatible route lacks. Set `max_output_tokens` generously (4096-8192 for long-form), turn `reasoning` off unless it is genuinely needed, keep `store: true` so a truncated answer can be continued with `previous_response_id`, and allow a long HTTP timeout for slow local models. ```python import os import requests LM_STUDIO_URL = "http://localhost:1234" TOKEN = os.environ.get("LM_API_TOKEN") # only needed when server auth is enabled HEADERS = {"Authorization": f"Bearer {TOKEN}"} if TOKEN else {} MODEL = os.environ.get("LM_MODEL", "local-worker") def _chat(**body) -> dict: body.setdefault("model", MODEL) body.setdefault("max_output_tokens", 4096) body.setdefault("reasoning", "off") # reasoning tokens eat the output budget body.setdefault("store", True) # required for previous_response_id r = requests.post(f"{LM_STUDIO_URL}/api/v1/chat", headers=HEADERS, json=body, timeout=900) # local models are slow; do not time out early r.raise_for_status() data = r.json() message = next(item for item in data["output"] if item["type"] == "message") return {"text": message["content"], "response_id": data.get("response_id"), "stats": data.get("stats", {})} def ask_local(prompt: str, system: str = "You are a helpful assistant.") -> dict: return _chat(input=prompt, system_prompt=system, temperature=0.3, context_length=32768) # match the loaded model's configured context def continue_local(response_id: str) -> dict: """Resume a cut-off answer without resending the source text.""" return _chat(previous_response_id=response_id, input="Continue exactly where you stopped. Do not repeat prior text.") result = ask_local("Summarize this text in 2 sentences: ...") print(result["text"]) # If stats["total_output_tokens"] reached max_output_tokens, the answer was cut off; extend it: # result = continue_local(result["response_id"]) ``` ### OpenAI-compatible fallback Use this route when the native API is unavailable (LM Studio older than 0.4.0, or an existing OpenAI-SDK codebase). Set `max_tokens` generously (4096-8192) because a small hard-coded limit silently truncates answers. There is no continuation primitive here, so always inspect `finish_reason`. ```python from openai import OpenAI # Point to local LM Studio server client = OpenAI( base_url="http://localhost:1234/v1", api_key="lm-studio", # any non-empty string; only checked when auth is on ) def ask_local(prompt: str, system: str = "You are a helpful assistant.") -> str: response = client.chat.completions.create( model="local-worker", # the identifier shown by /v1/models messages=[ {"role": "system", "content": system}, {"role": "user", "content": prompt}, ], temperature=0.3, max_tokens=4096, # generous; check finish_reason below ) if response.choices[0].finish_reason == "length": raise RuntimeError("LM Studio truncated the response; use the native API continuation flow.") return response.choices[0].message.content # Example usage result = ask_local("Summarize this text in 2 sentences: ...") print(result) ``` ### Task C: Create task-specific subagents ```python from openai import OpenAI client = OpenAI(base_url="http://localhost:1234/v1", api_key="lm-studio") class LocalSubagent: def __init__(self, system_prompt: str, temperature: float = 0.2): self.system_prompt = system_prompt self.temperature = temperature def run(self, user_input: str) -> str: response = client.chat.completions.create( model="local-worker", messages=[ {"role": "system", "content": self.system_prompt}, {"role": "user", "content": user_input}, ], temperature=self.temperature, max_tokens=4096, # generous; check finish_reason below ) if response.choices[0].finish_reason == "length": raise RuntimeError("Response truncated; retry through the native API continuation flow.") return response.choices[0].message.content # Define specialized subagents summarizer = LocalSubagent( system_prompt="You are a summarization expert. Produce concise 2-3 sentence summaries." ) classifier = LocalSubagent( system_prompt="Classify the input into one of these categories: billing, technical, general, urgent. Respond with only the category name.", temperature=0.0, ) extractor = LocalSubagent( system_prompt="Extract all named entities (people, organizations, dates, amounts) from the text. Return as JSON.", temperature=0.0, ) # Use the subagents summary = summarizer.run("Long document text here...") category = classifier.run("I can't log into my account and I need to submit a report by EOD") entities = extractor.run("John Smith signed a $50,000 contract with Acme Corp on March 15, 2025") ``` ### Task D: Batch processing with local models ```python import asyncio from openai import AsyncOpenAI client = AsyncOpenAI(base_url="http://localhost:1234/v1", api_key="lm-studio") async def process_batch(items: list[str], system_prompt: str, max_concurrent: int = 4, max_tokens: int = 4096) -> list[dict]: semaphore = asyncio.Semaphore(max_concurrent) async def process_one(text: str) -> dict: async with semaphore: response = await client.chat.completions.create( model="local-worker", messages=[ {"role": "system", "content": system_prompt}, {"role": "user", "content": text}, ], temperature=0.2, max_tokens=max_tokens, # generous but finite: unbounded batches can run for hours ) choice = response.choices[0] # Flag truncation per item instead of raising — one bad item must not # discard the other 99 results. return {"text": choice.message.content, "truncated": choice.finish_reason == "length"} tasks = [process_one(item) for item in items] results = await asyncio.gather(*tasks, return_exceptions=True) return [r if isinstance(r, dict) else {"text": None, "error": str(r)} for r in results] # Batch summarize 100 documents documents = ["doc1 text...", "doc2 text...", ...] # 100 documents summaries = asyncio.run(process_batch( documents, system_prompt="Summarize in 2 sentences.", max_concurrent=2, # keep low on a single local GPU )) incomplete = [i for i, r in enumerate(summaries) if r.get("truncated") or r.get("error")] ``` ### Task E: Cost comparison and routing strategy Decide when to use local vs. cloud models: | Task | Local Model | Cloud API | Recommendation | |------|------------|-----------|----------------| | Summarization | Good | Better | Local (save cost) | | Classification | Good | Good | Local (save cost) | | Data extraction | Moderate | Good | Local for simple, cloud for complex | | Code generation | Moderate | Better | Cloud | | Complex reasoning | Weak | Strong | Cloud | | Translation | Good | Better | Local for common languages | ```python def smart_route(task_type: str, text: str) -> str: """Route tasks between local and cloud models.""" local_tasks = {"summarize", "classify", "extract_simple", "reformat"} if task_type in local_tasks: return ask_local(text) # Free, local inference else: return ask_cloud(text) # Paid, cloud API ``` ## Examples ### Example 1: Summarize 500 support tickets locally **User request:** "Summarize all our support tickets from last month without API costs" ```python tickets = load_tickets_from_csv("tickets.csv") summaries = asyncio.run(process_batch( [t["description"] for t in tickets], system_prompt="Summarize this support ticket in one sentence. Include the main issue and any resolution.", max_concurrent=2, )) # Re-run any item where r["truncated"] is True through the native continuation flow. # Cost: $0 (vs ~$15 with GPT-4) ``` ### Example 2: Classify incoming emails **User request:** "Auto-classify emails into categories using a local model" ```python classifier = LocalSubagent( system_prompt="Classify this email into exactly one category: sales, support, spam, internal. Reply with only the category.", temperature=0.0, ) for email in emails: category = classifier.run(email["subject"] + "\n" + email["body"]) email["category"] = category.strip().lower() ``` ### Example 3: Extract structured data from documents **User request:** "Extract names, dates, and amounts from these contracts" ```python extractor = LocalSubagent( system_prompt='Extract fields from the contract as JSON: {"parties": [], "date": "", "amount": "", "term": ""}', temperature=0.0, ) for doc in contracts: data = extractor.run(doc["text"]) print(f"{doc['filename']}: {data}") ``` ## Guidelines - Local hardware is the bottleneck: start batch jobs with `max_concurrent=1-2` and raise it only if tokens-per-second (in `stats`) holds up. - Prefer the native `/api/v1/chat` endpoint (LM Studio 0.4.0+) for long-form requests; use `/v1/chat/completions` as a compatibility fallback. - Never hard-code small output limits such as `512`, `1024`, or `2048` for long-form tasks — that is the usual cause of summaries that stop mid-sentence. Use `max_output_tokens` of 4096-8192 natively, or `max_tokens` of 4096-8192 on the compatible route. - Never present a length-terminated response as complete: check `finish_reason == "length"` on the compatible route and compare `stats.total_output_tokens` with `max_output_tokens` natively, then resume with `previous_response_id` rather than resending the source. - Use `reasoning: "off"` for summaries and extraction; reasoning tokens consume the generation budget without appearing in the returned text. - Context length and output length are separate limits — raise `context_length` at model-load time for large inputs, and `max_output_tokens` for long answers. - Use quantized models (Q4_K_M or Q5_K_M) for best speed-to-quality ratio on consumer hardware. - 8B parameter models are the sweet spot for most extraction and classification tasks. - Set `temperature=0.0` for deterministic tasks like classification and extraction. - Test local model accuracy on a sample of 20-50 items before running full batches. - For tasks where local models underperform, fall back to cloud APIs automatically. - Keep LM Studio running as a background service for always-on local inference. - Monitor RAM and VRAM usage; 7B models need ~6 GB RAM (quantized) or ~16 GB (full precision). --- Source attribution: terminal-skills Original project: https://github.com/TerminalSkills/skills/blob/8a9852d489f274538be9bf1c9121aeb403750b1c/skills/lm-studio-subagents/SKILL.md License: Apache-2.0 License information: https://www.apache.org/licenses/LICENSE-2.0 Original upstream instructions; downloaded from Dave AI. ## Upstream license notice Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 1. Definitions. "License" shall mean the terms and conditions for use, reproduction, and distribution as defined by Sections 1 through 9 of this document. "Licensor" shall mean the copyright owner or entity authorized by the copyright owner that is granting the License. "Legal Entity" shall mean the union of the acting entity and all other entities that control, are controlled by, or are under common control with that entity. For the purposes of this definition, "control" means (i) the power, direct or indirect, to cause the direction or management of such entity, whether by contract or otherwise, or (ii) ownership of fifty percent (50%) or more of the outstanding shares, or (iii) beneficial ownership of such entity. "You" (or "Your") shall mean an individual or Legal Entity exercising permissions granted by this License. "Source" form shall mean the preferred form for making modifications, including but not limited to software source code, documentation source, and configuration files. "Object" form shall mean any form resulting from mechanical transformation or translation of a Source form, including but not limited to compiled object code, generated documentation, and conversions to other media types. "Work" shall mean the work of authorship, whether in Source or Object form, made available under the License, as indicated by a copyright notice that is included in or attached to the work (an example is provided in the Appendix below). "Derivative Works" shall mean any work, whether in Source or Object form, that is based on (or derived from) the Work and for which the editorial revisions, annotations, elaborations, or other modifications represent, as a whole, an original work of authorship. For the purposes of this License, Derivative Works shall not include works that remain separable from, or merely link (or bind by name) to the interfaces of, the Work and Derivative Works thereof. "Contribution" shall mean any work of authorship, including the original version of the Work and any modifications or additions to that Work or Derivative Works thereof, that is intentionally submitted to the Licensor for inclusion in the Work by the copyright owner or by an individual or Legal Entity authorized to submit on behalf of the copyright owner. For the purposes of this definition, "submitted" means any form of electronic, verbal, or written communication sent to the Licensor or its representatives, including but not limited to communication on electronic mailing lists, source code control systems, and issue tracking systems that are managed by, or on behalf of, the Licensor for the purpose of discussing and improving the Work, but excluding communication that is conspicuously marked or otherwise designated in writing by the copyright owner as "Not a Contribution." "Contributor" shall mean Licensor and any individual or Legal Entity on behalf of whom a Contribution has been received by the Licensor and subsequently incorporated within the Work. 2. Grant of Copyright License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form. 3. Grant of Patent License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable (except as stated in this section) patent license to make, have made, use, offer to sell, sell, import, and otherwise transfer the Work, where such license applies only to those patent claims licensable by such Contributor that are necessarily infringed by their Contribution(s) alone or by combination of their Contribution(s) with the Work to which such Contribution(s) was submitted. If You institute patent litigation against any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Work or a Contribution incorporated within the Work constitutes direct or contributory patent infringement, then any patent licenses granted to You under this License for that Work shall terminate as of the date such litigation is filed. 4. Redistribution. You may reproduce and distribute copies of the Work or Derivative Works thereof in any medium, with or without modifications, and in Source or Object form, provided that You meet the following conditions: (a) You must give any other recipients of the Work or Derivative Works a copy of this License; and (b) You must cause any modified files to carry prominent notices stating that You changed the files; and (c) You must retain, in the Source form of any Derivative Works that You distribute, all copyright, patent, trademark, and attribution notices from the Source form of the Work, excluding those notices that do not pertain to any part of the Derivative Works; and (d) If the Work includes a "NOTICE" text file as part of its distribution, then any Derivative Works that You distribute must include a readable copy of the attribution notices contained within such NOTICE file, excluding any notices that do not pertain to any part of the Derivative Works, in at least one of the following places: within a NOTICE text file distributed as part of the Derivative Works; within the Source form or documentation, if provided along with the Derivative Works; or, within a display generated by the Derivative Works, if and wherever such third-party notices normally appear. The contents of the NOTICE file are for informational purposes only and do not modify the License. You may add Your own attribution notices within Derivative Works that You distribute, alongside or as an addendum to the NOTICE text from the Work, provided that such additional attribution notices cannot be construed as modifying the License. You may add Your own copyright statement to Your modifications and may provide additional or different license terms and conditions for use, reproduction, or distribution of Your modifications, or for any such Derivative Works as a whole, provided Your use, reproduction, and distribution of the Work otherwise complies with the conditions stated in this License. 5. Submission of Contributions. Unless You explicitly state otherwise, any Contribution intentionally submitted for inclusion in the Work by You to the Licensor shall be under the terms and conditions of this License, without any additional terms or conditions. Notwithstanding the above, nothing herein shall supersede or modify the terms of any separate license agreement you may have executed with Licensor regarding such Contributions. 6. Trademarks. This License does not grant permission to use the trade names, trademarks, service marks, or product names of the Licensor, except as required for reasonable and customary use in describing the origin of the Work and reproducing the content of the NOTICE file. 7. Disclaimer of Warranty. Unless required by applicable law or agreed to in writing, Licensor provides the Work (and each Contributor provides its Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, including, without limitation, any warranties or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. You are solely responsible for determining the appropriateness of using or redistributing the Work and assume any risks associated with Your exercise of permissions under this License. 8. Limitation of Liability. In no event and under no legal theory, whether in tort (including negligence), contract, or otherwise, unless required by applicable law (such as deliberate and grossly negligent acts) or agreed to in writing, shall any Contributor be liable to You for damages, including any direct, indirect, special, incidental, or consequential damages of any character arising as a result of this License or out of the use or inability to use the Work (including but not limited to damages for loss of goodwill, work stoppage, computer failure or malfunction, or any and all other commercial damages or losses), even if such Contributor has been advised of the possibility of such damages. 9. Accepting Warranty or Additional Liability. While redistributing the Work or Derivative Works thereof, You may choose to offer, and charge a fee for, acceptance of support, warranty, indemnity, or other liability obligations and/or rights consistent with this License. However, in accepting such obligations, You may act only on Your own behalf and on Your sole responsibility, not on behalf of any other Contributor, and only if You agree to indemnify, defend, and hold each Contributor harmless for any liability incurred by, or claims asserted against, such Contributor by reason of your accepting any such warranty or additional liability. END OF TERMS AND CONDITIONS Copyright 2025 Terminal Skills Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. # Dave AI resource metadata Creator / publisher: terminal-skills Original project: https://github.com/TerminalSkills/skills/blob/8a9852d489f274538be9bf1c9121aeb403750b1c/skills/lm-studio-subagents/SKILL.md Attribution basis: Source repository record Fields of work: Development & Engineering, Data & Analytics Roles: ML / AI Engineer, Data Scientist, Data Engineer Specialties: Frontend, Word Documents, API, Testing, Backend Tools: Lm Studio Country applicability: Country not specified Classification basis: Source category, role and specialization labels Instruction SHA-256: 409d869d363bd0f8aa99f00cd7b09797173f4c2fc2431dcb43b0b49afdb0552c