{"slug": "botpsy-0-8b-tether-s-open-on-device-agent-model-for-tool-calling", "title": "BotPsy 0.8B: Tether's Open On-Device Agent Model for Tool Calling", "summary": "Tether AI Research on November 11, 2026 published BotPsy, a 0.8B-parameter on-device agent model built on Qwen3.5-0.8B that targets function calling and multi-step tool workflows on phones and laptops under an Apache 2.0 license for research and education. Tether reports BotPsy roughly doubles base tool-use capability, scoring 44.0 on BFCL v4 versus 23.4 for the Qwen3.5-0.8B base and 37.7 for Qwen3.5-2B-Think, and leading the ≤1.2B class on five of six benchmarks including ACEBench (58.7), τ-bench (51.5) and ToolSandbox (64.6). Tether also reports a 21% unsafe rate in its red-team harness versus 36% for the base model, and ships Safetensors and GGUF (bf16, Q8, Q4) weights for QVAC and llama.cpp.", "body_md": "**On November 11, 2026 Tether AI Research published [BotPsy](https://huggingface.co/blog/qvac/botpsy?ref=explainx), a **0.8B-parameter** agent model built on **Qwen3.5-0.8B** that targets **function calling** and **multi-step tool workflows** on **phones and laptops**—with **Apache 2.0** weights for research and education.** The [Hugging Face post](https://huggingface.co/blog/qvac/botpsy?ref=explainx) argues many routines do not need frontier cloud models, and that **sub-billion** agents can be **cheaper**, **more private**, and **less hostage to API policy changes**.\n\nIf you are shipping **local agents**—calendar helpers, device automations, or personal research bots—BotPsy is one of the few **sub-1B** releases that publishes **six agent benchmarks** and a **full training recipe**, not only a merge on top of a base chat model.\n\n## TL;DR: BotPsy at a glance\n\n| Question | Answer (per Tether's Nov 11, 2026 post) | \n|---|---|\n| Base model | **Qwen3.5-0.8B** | \n| Size class | **0.8B** parameters | \n| License | **Apache 2.0** (research and education) | \n| Formats | **Safetensors** +**GGUF** (bf16, Q8, Q4) for QVAC /**llama.cpp** | \n| Headline claim | **~2x** agentic scores vs base on BFCL v4, ACEBench, τ-bench | \n| Vs 2B sibling | Beats **Qwen3.5-2B-Think** on**5/6** tool benchmarks in their table | \n| Safety | **21%** unsafe rate in their red-team harness vs**36%** for base | \n| Intended use | **Low–medium** difficulty personal assistant tasks—not high stakes | \n\nNumbers are **author-reported** unless you reproduce the eval harness yourself.\n\n## Why another sub-billion agent model matters\n\nCloud agents are easy to prototype and expensive to operate at scale. A personal assistant that **searches**, **reads mail**, or **calls home APIs** can burn **millions of tokens** if every step hits a frontier endpoint. Small models also reduce **vendor lock-in**: the Hugging Face post explicitly lists **deprecation risk** and **limited customization** as reasons to host locally.\n\nTether already ships on-device speech with [TranscribePsy](https://explainx.ai/blog/tether-transcribepsy-250m-on-device-speech-recognition-2026). BotPsy extends the same **QVAC** story from **audio** to **tool-use agents**, sitting beside other **local** stacks covered in [Perplexity Computer running locally on a Mac](https://explainx.ai/blog/perplexity-computer-fully-local-mac-qwen-3-6-pplx-pro-36gb-offline-what-stays-on-device-2026) and [Liquid on-device reasoning models](https://explainx.ai/blog/liquid-ai-lfm2-5-24b-a2b-open-weight-on-device-reasoning-agent-model-2026).\n\nLocal on-device agent language model running tool calls without cloud API dependency\n\n## Benchmarks: what BotPsy claims to beat\n\n### Tool-use and agent suites\n\nTether compares **BotPsy-0.8B** against **Qwen3.5-0.8B-Think** (base), **Qwen3.5-2B-Think**, **LFM2.5-1.2B-Think**, and **MiniCPM5-1B-Think** on six benchmarks: **BFCL v4**, **ACEBench**, **τ-bench**, **Claw-Eval**, **BrowseComp-Plus**, and **ToolSandbox**.\n\nReported highlights from the blog:\n\n| Benchmark | BotPsy-0.8B | Qwen3.5-0.8B base | Qwen3.5-2B-Think | \n|---|---|---|---|\n| BFCL v4 | **44.0** | 23.4 | 37.7 | \n| ACEBench | **58.7** | 30.1 | 55.4 | \n| τ-bench | **51.5** | 23.2 | 46.5 | \n| ToolSandbox | **64.6** | 30.5 | 41.2 | \n| Claw-Eval | 46.5 | 39.3 | **48.9** | \n| BrowseComp-Plus | **4.5** | 1.9 | 2.8 | \n\nThe blog states BotPsy **roughly doubles** base tool-use capability and leads the **≤1.2B** class on **five of six** suites.\n\nBotPsy 0.8B on-device agent model tool-use benchmark scores versus sub-1.2B rivals\n*Image: [Tether AI Research / Hugging Face](https://huggingface.co/blog/qvac/botpsy?ref=explainx)*\n\n### Multi-turn customer-service style tasks\n\nOn **τ²-bench telecom**, BotPsy reports **82.5** vs **69.3** for Qwen3.5-2B-Think. On **BFCL v4 multi-turn**, it rises from **7.5** (base) to **46.1**, which the authors contrast with **23.8** for the 2B model.\n\nThat matters because **multi-turn** tool loops are where tiny models usually collapse—either they **forget policy** or they **call tools with drifting arguments**. The results rhyme with why enterprises benchmark assistants on [τ²-style benches](https://explainx.ai/blog/raindrop-reasoning-agent-memory-tau2-bench-67-percent-2026) instead of single-shot trivia.\n\n### Knowing when **not** to call tools\n\nA capable agent must **decline** when no tool applies. On **BFCL v4 hallucination tests**, BotPsy scores **86.1**, the highest among models tested in the post. On **ACEBench special cases** (incomplete requests, wrong parameters), it scores **51.3** vs **13.7** for the base model.\n\nThat behavior is underrated in hobby agent demos but decisive in production—see [eBay blocking third-party shopping agents](https://explainx.ai/blog/ebay-blocks-third-party-ai-agents-browsing-shopping-2026) for how **uncontrolled tool loops** create platform risk.\n\n### Knowledge vs tool specialization\n\nTool training did not erase knowledge entirely: **GPQA Diamond** moves from **33.2** to **38.1** and **MMLU-Pro** from **41.6** to **48.9** in the authors' tables, best among **≤1.2B** models tested. The blog still warns that **knowledge scales with parameters** and BotPsy should **call tools** for facts rather than act as a standalone knowledge model—similar philosophy to [Hugging Face System One decision APIs](https://explainx.ai/blog/hugging-face-decision-models-systemone-api-2026) that push **structured decisions** instead of long free text.\n\n## Training recipe: four stages, three feedback types\n\nBotPsy is not a single SFT pass. The [Hugging Face blog](https://huggingface.co/blog/qvac/botpsy?ref=explainx) describes:\n\n1. **Supervised fine-tuning** on multi-turn tool trajectories, parallel and multi-step calls, and**no-tool** cases.\n2. **Multi-teacher on-policy distillation** from three**Qwen3.5-4B** specialists (tool use, reasoning, instruction following) with per-token reverse KL.\n3. **RL round 1** with**DAPO** and fine-grained tool-call rewards inspired by ToolRL-style partial credit.\n4. **RL round 2** mixing math, code, knowledge MCQs, instruction following, structured outputs, and simulated workplace tool use.\n\nThe post includes stage-by-stage capability curves showing **SFT boosts tools but hurts knowledge**, then **distillation recovers** general skills before RL adds further gains—useful reading if you are building your own [agent skills](https://explainx.ai/blog/what-are-agent-skills-complete-guide) distillation pipeline.\n\n## Safety, efficiency, and deployment notes\n\n### Red-teaming numbers\n\nAutomated red-teaming across **six harm categories** reports **21%** unsafe responses for BotPsy vs **36%** for its base and **35–47%** for some 1B-class rivals in the same table. Under **prompt-injection** attacks tied to untrusted tool output, unsafe rate falls from **46%** (base) to **31%**.\n\nThose rates are still **too high** for unsupervised customer support. The blog's own **Intended Use** section says BotPsy is **not** for high-stakes use.\n\n### Reasoning token efficiency\n\nOn questions all five models answer correctly, BotPsy uses about **3× fewer reasoning tokens** than base on GPQA Diamond and **2.7× fewer** on MMLU-Pro, with loop rates collapsing on IFBench in their charts. For on-device latency, **shorter chains** matter as much as raw accuracy.\n\n### How to try BotPsy\n\nThe post links **Hugging Face safetensors** and **GGUF** quantizations for **QVAC SDK** and [**llama.cpp**](https://github.com/ggml-org/llama.cpp?ref=explainx). Typical local stack:\n\n1. Pull weights from the model card linked in the blog.\n2. Pair with a **host app** that exposes tools (calendar, search, files) with explicit schemas.\n3. Add **policy prompts** and**human confirmation** for irreversible actions—the model will not supply enterprise governance for you.\n\nCompare setup friction with [Janestreet's agent-skills tutor repo](https://explainx.ai/blog/janestreet-agent-skills-socratic-tutor-github-2026) if you want a skills-first harness instead of a monolithic assistant binary.\n\n## What Tether says comes next\n\nThe vision section mentions **Hypersearch** P2P queries, **QVAC Fabric** local fine-tuning, delegating heavy compute to home PCs, and **agent-to-agent** messaging on trusted LANs—all **on-device control** themes. Treat that as **roadmap prose**, not shipped product.\n\n## Hardware and runtime expectations\n\nTether positions BotPsy for **phone-class** deployment via **GGUF** quantizations. Realistic planning:\n\n| Precision | Likely role | Caveat | \n|---|---|---|\n| **bf16** | Laptop prototyping | Higher RAM, best tool-call fidelity in author tests | \n| **Q8** | Daily-driver phone experiments | Watch thermals on long multi-turn loops | \n| **Q4** | Offline demos | Expect JSON/schema errors to rise; revalidate tools | \n\nYou still need a **host process** implementing schemas—same lesson as [Expo Exact contract runtime](https://explainx.ai/blog/expo-exact-contract-agent-native-cross-platform-runtime-2026) posts about **typed tool surfaces**. BotPsy does not ship a calendar, browser, or payments stack.\n\nOn Apple silicon, compare latency against [Liquid's on-device reasoning line](https://explainx.ai/blog/liquid-ai-lfm2-5-24b-a2b-open-weight-on-device-reasoning-agent-model-2026): BotPsy is **smaller** but **more tool-specialized**.\n\n## Benchmarks readers should rerun before betting products\n\nTether lists six suites—each measures a different failure mode:\n\n- **BFCL v4** — schema adherence and multi-turn calls.\n- **ACEBench** — messy user phrasing and special cases.\n- **τ-bench / τ²-bench** — long customer-service dialogues with policy.\n- **Claw-Eval** — claw-style tool environments where 2B still wins slightly.\n- **BrowseComp-Plus** — web research (everyone scores near zero).\n- **ToolSandbox** — simulated app integrations.\n\nCopy the authors' lesson: **optimize the benchmark that matches your app group**—commerce vs device control vs messaging—not the unweighted mean. Figure 7 in the blog breaks scores by **app group** for simple vs complex scenarios; mobile assistants heavy on **device settings** should study that slice before trusting the headline average.\n\n## Open weights and licensing nuance\n\nApache **2.0** for research and education is broader than many \"open\" models that block distillation or commercial fine-tunes. Still read Tether's **Intended Use** guardrails: they flag **high-stakes** automation and document **21%** unsafe generations in their harness.\n\nIf you bundle BotPsy into a consumer app, your privacy policy must state **on-device** inference clearly—regulators already scrutinize **default cloud** assistants. Pair with [Hugging Face's open-source security tradeoff essay](https://explainx.ai/blog/hugging-face-open-source-ai-security-tradeoff-solaiman-2026) when security asks why you chose local weights.\n\n## What people are asking\n\n| Question | Honest answer | \n|---|---|\n| Is BotPsy better than Claude or GPT for agents? | Not on hard tasks. It targets **low–medium** personal workflows at**sub-billion** cost. | \n| Can I fine-tune BotPsy on my mail? | Apache weights allow research fine-tunes; check the license scope and your privacy policy. | \n| Does BotPsy include tools? | **No.** It is weights plus eval claims; you build the tool host. | \n| How does it relate to Qwen3.5? | It **is**[Qwen3.5-0.8B](https://huggingface.co/Qwen/Qwen3.5-0.8B?ref=explainx) plus Tether's four-stage post-training. | \n\n## Limitations you should plan for\n\n- **BrowseComp-Plus** scores are**single digits** for everyone—web research is still hard at this scale.\n- **Lower-resource languages** lag on instruction following per the blog's Figure 2 discussion.\n- **Banking-style τ³** tasks stay near zero for all models tested—do not point BotPsy at treasury workflows.\n- Benchmarks are **internal reproductions** ; wait for community runs before betting a product roadmap on the table.\n\n## Related reading on explainx.ai\n\n*Model names, scores, and release details follow Tether's November 11, 2026 Hugging Face post.*", "url": "https://wpnews.pro/news/botpsy-0-8b-tether-s-open-on-device-agent-model-for-tool-calling", "canonical_source": "https://explainx.ai/blog/botpsy-0-8b-on-device-agent-qvac-apache-tool-use-2026?utm_source=rss&utm_medium=feed&utm_campaign=blog", "published_at": "2026-11-11 00:00:00+00:00", "updated_at": "2026-11-11 08:51:39.049476+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "large-language-models", "ai-research", "ai-safety"], "entities": ["Tether AI Research", "BotPsy", "Qwen3.5-0.8B", "Qwen3.5-2B-Think", "Hugging Face", "QVAC", "llama.cpp", "BFCL v4"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/botpsy-0-8b-tether-s-open-on-device-agent-model-for-tool-calling", "markdown": "https://wpnews.pro/news/botpsy-0-8b-tether-s-open-on-device-agent-model-for-tool-calling.md", "text": "https://wpnews.pro/news/botpsy-0-8b-tether-s-open-on-device-agent-model-for-tool-calling.txt", "jsonld": "https://wpnews.pro/news/botpsy-0-8b-tether-s-open-on-device-agent-model-for-tool-calling.jsonld"}}