{"id":50578257,"url":"https://github.com/owllm/owllm","last_synced_at":"2026-06-26T06:01:04.727Z","repository":{"id":362281175,"uuid":"1258198877","full_name":"OwLLM/owllm","owner":"OwLLM","description":"Local-first AI workstation. Run open-weight models, fine-tune, orchestrate multi-agent teams. No cloud required.","archived":false,"fork":false,"pushed_at":"2026-06-16T12:56:46.000Z","size":278,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-16T14:18:06.297Z","etag":null,"topics":["abliteration","agentic","ai-agents","desktop-app","fine-tuning","huggingface","llama-cpp","llm","local-first","local-llm","mcp","model-context-protocol","multi-agent","privacy","tauri","unsloth"],"latest_commit_sha":null,"homepage":null,"language":null,"has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"other","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/OwLLM.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":"CODE_OF_CONDUCT.md","threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":"SECURITY.md","support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-03T11:07:32.000Z","updated_at":"2026-06-14T03:44:45.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/OwLLM/owllm","commit_stats":null,"previous_names":["owllm/owllm"],"tags_count":120,"template":false,"template_full_name":null,"purl":"pkg:github/OwLLM/owllm","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OwLLM%2Fowllm","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OwLLM%2Fowllm/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OwLLM%2Fowllm/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OwLLM%2Fowllm/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/OwLLM","download_url":"https://codeload.github.com/OwLLM/owllm/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OwLLM%2Fowllm/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34805072,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-26T02:00:06.560Z","response_time":106,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["abliteration","agentic","ai-agents","desktop-app","fine-tuning","huggingface","llama-cpp","llm","local-first","local-llm","mcp","model-context-protocol","multi-agent","privacy","tauri","unsloth"],"created_at":"2026-06-05T00:01:19.828Z","updated_at":"2026-06-26T06:01:04.709Z","avatar_url":"https://github.com/OwLLM.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\r\n\r\n\u003cimg src=\"assets/OWLLM_Hero.png\" alt=\"OwLLM\" width=\"600\"\u003e\r\n\r\n# ⬇ [Download OwLLM for Windows](https://github.com/OwLLM/owllm/releases/latest/download/OwLLM.Desktop.Setup.exe)\r\n\r\n### One file · ~30 MB · No admin required · Windows 10 / 11 x64\r\n\r\n[![installer](https://img.shields.io/badge/installer-OwLLM.Desktop.Setup.exe-3ec5d8?style=for-the-badge\u0026logo=windows)](https://github.com/OwLLM/owllm/releases/latest/download/OwLLM.Desktop.Setup.exe)\r\n[![portable](https://img.shields.io/badge/portable_(zip)-no_install-7e8aa0?style=for-the-badge)](https://github.com/OwLLM/owllm/releases/latest/download/OwLLM-Desktop-Portable.zip)\r\n\r\n---\r\n\r\n### Your team of AI agents. Build them. Own them. Run them anywhere.\r\n\r\n**OwLLM is an open platform to build, deploy, and run custom AI agent teams — on your hardware, your VPS, or in a VM, 24/7. Bring your own models: local, cloud, or both. Fine-tune. Quantize. Abliterate. Red-team. Automate.**\r\n\r\n\u003cp align=\"center\"\u003e\r\n  \u003ca href=\"https://github.com/OwLLM/owllm/releases/latest\"\u003e\u003cimg src=\"https://img.shields.io/github/v/release/OwLLM/owllm?label=version\u0026style=for-the-badge\u0026color=3ec5d8\" alt=\"Latest version\" /\u003e\u003c/a\u003e\r\n  \u003ca href=\"https://github.com/OwLLM/owllm/discussions\"\u003e\u003cimg src=\"https://img.shields.io/github/discussions/OwLLM/owllm?style=for-the-badge\u0026color=3ec5d8\" alt=\"Discussions\" /\u003e\u003c/a\u003e\r\n  \u003ca href=\"LICENSE\"\u003e\u003cimg src=\"https://img.shields.io/github/license/OwLLM/owllm?style=for-the-badge\u0026color=3ec5d8\" alt=\"License\" /\u003e\u003c/a\u003e\r\n  \u003ca href=\"https://github.com/OwLLM/owllm/stargazers\"\u003e\u003cimg src=\"https://img.shields.io/github/stars/OwLLM/owllm?style=for-the-badge\u0026color=3ec5d8\" alt=\"Stars\" /\u003e\u003c/a\u003e\r\n\u003c/p\u003e\r\n\r\n\u003e [!IMPORTANT]\r\n\u003e **OwLLM Desktop currently ships for Windows 10/11 (x64) only.** macOS (Apple Silicon + Intel) and Linux (x86_64) builds are coming via the already-configured cross-platform CI. Watch the repo for release notifications.\r\n\r\n\u003c/div\u003e\r\n\r\n---\r\n\r\n## What makes OwLLM different\r\n\r\nMost AI tools give you a chatbox. **OwLLM gives you a workforce.**\r\n\r\nYou compose teams of specialised agents — an orchestrator that plans, a coder that writes, a critic that reviews, a researcher that fact-checks — and they collaborate on real tasks in parallel. Each team is a graph of roles + prompts you define. The 18 teams shipped in this repo are **starter samples**, not the menu.\r\n\r\n| | What OwLLM gives you that others don't |\r\n|---|---|\r\n| 🧩 **Build your own teams** | Compose agents from 8 base roles + custom prompts. Visual graph builder. Hot-updates through this repo — push a team JSON, it lands on every installed app. |\r\n| ☁️ **Cloud OR local — same teams** | No 4090? Plug in Claude / GPT / Gemini / Kimi keys, teams work identically. Have a GPU? Run open-weight models locally and stop paying per token. Mix both in the same conversation. |\r\n| 🎓 **Fine-tune any model** | Full LoRA + Unsloth + TRL pipeline. Drop a JSONL, watch loss curves, save adapters. Works on consumer GPUs (8 GB+). |\r\n| 🔬 **Abliterate for safety research** | Orthogonalise weights against refusal directions. Generate adversarial datasets. Train better safety classifiers. The honest tools the field actually needs. |\r\n| 🛠 **GGUF + quantization built-in** | Convert HF safetensors → GGUF, quantize Q4/Q5/Q6/Q8/F16. Ship custom models anyone with llama.cpp can run. |\r\n| 🛡 **Red-team capable** | Compose adversarial agent teams whose *job* is to find vulnerabilities — in models, code, apps. Pair with fine-tuning to train defenders. |\r\n| 🔒 **Sealed to ONE folder (Win/Mac/Linux)** | Flip it on and every tool your agents run — shell, file writes, edits, search, **and the cloud CLIs (Claude/Codex/Gemini/Kimi)** — runs **inside a real Linux sandbox** (WSL2 on Windows, Lima VM on macOS, bubblewrap on Linux *(Mac/Linux beta)*). Agents are **sealed to ONLY the project folder**: they work on your **real Windows folder — no copy** — but **cannot see the rest of your C: drive, other projects, or your home / SSH keys**. A model that runs `rm -rf` simply has nothing else to reach. Isolated **by default**; the seal and toolchain **auto-install**; provider logins (CLIs **and** API keys) **auto-sync** in so cloud agents just work. **Connect GitHub** to clone/push private repos from inside. See the **🔒 Sealed to one folder** section below for the diagram. |\r\n| 🔌 **MCP-first tooling** | Plug in any Model Context Protocol server (filesystem, git, browser, Postgres, GitHub…). **Keyless DuckDuckGo web search is auto-installed on first run** — no API key, no card. Engine-agnostic: any search MCP you add is used automatically. Curated packs per team. |\r\n| 🏠 **Run anywhere** | Desktop today. **Headless on a $5/mo VPS, 24/7** — on the roadmap. Containerised / VM — on the roadmap. Your agents, your hardware, your terms. |\r\n\r\n## 🔒 Sealed to one folder — nothing else\r\n\r\nGive an AI agent a shell and it can read your SSH keys, wander into other repos, or `rm -rf` the wrong thing. OwLLM puts a hard wall around that. Turn isolation on (it's the default) and every command runs inside a Linux sandbox that can see **only the project folder you're working on** — your **real** Windows folder, mounted live with **no copy** — and nothing else of your machine.\r\n\r\n```mermaid\r\nflowchart LR\r\n    A[\"🤖 Agent\u003cbr/\u003eshell · edits · cloud CLIs\u003cbr/\u003e(Claude · Codex · Gemini)\"]\r\n    A ==\u003e|\"sealed by bubblewrap inside WSL2\"| BOX\r\n\r\n    subgraph BOX[\"🔒 The sandbox sees ONLY this\"]\r\n        direction TB\r\n        P[\"📁 Your project folder\u003cbr/\u003eyour real Windows files — no copy\"]\r\n        K[\"🔑 The agent's own logins\u003cbr/\u003eauto-synced — cloud agents just work\"]\r\n    end\r\n\r\n    BOX -.-\u003e|\"🚫 blocked\"| C1[\"🗂️ Rest of your C: drive\"]\r\n    BOX -.-\u003e|\"🚫 blocked\"| C2[\"📦 Your other projects\"]\r\n    BOX -.-\u003e|\"🚫 blocked\"| C3[\"🔐 Home · SSH keys · browser\"]\r\n\r\n    classDef safe fill:#0e2a1e,stroke:#3ec58a,stroke-width:2px,color:#d6ffe9;\r\n    classDef block fill:#2a1414,stroke:#d86a6a,stroke-width:1px,color:#ffd9d9;\r\n    classDef agent fill:#0a2230,stroke:#3ec5d8,stroke-width:2px,color:#dff6ff;\r\n    class A agent;\r\n    class P,K safe;\r\n    class C1,C2,C3 block;\r\n    style BOX fill:#0c1f17,stroke:#3ec58a,stroke-width:2px,color:#d6ffe9;\r\n```\r\n\r\n- **Real folder, no copy** — work on `C:\\code\\my-repo` directly; changes are immediate, nothing is duplicated, no disk doubling.\r\n- **Keeps working after you seal it** — the agent's Claude/Codex/Gemini logins are bound in, so the team doesn't suddenly ask you to log in again.\r\n- **One toggle, honest feedback** — press **🔍 Verify** and the app runs a probe through the agent's *own* shell and tells you plainly: *runs in WSL? sealed to only this folder?* — and installs the seal if it's missing.\r\n\r\n### 💾 …and it cleans up after itself\r\n\r\nThe sandbox lives in WSL, whose virtual disk only ever grows. A **Sandbox disk** card on the Home page keeps it honest:\r\n\r\n| Action | What it does | Cost |\r\n|---|---|:---:|\r\n| **See usage** | WSL disk size + reclaimable caches (uv / npm / pip) + project-copy size, at a glance | — |\r\n| **🧹 Clear caches** | Drops regenerable build caches — never your projects, logins, or models. Often frees **many GB** in seconds | instant, no restart |\r\n| **💿 Reclaim disk** | Physically **shrinks** the WSL disk file and reports exactly how much came back | restarts WSL · admin prompt |\r\n\r\nDeleting a project also frees its sandbox copy automatically — your own folders are never touched.\r\n\r\n## What teams can do\r\n\r\nOwLLM ships starter teams in nine categories. **All of them are forkable and remixable** — they're templates, not the menu. The real product is the team builder.\r\n\r\n| Category | What teams here do | Starter samples |\r\n|---|---|---|\r\n| 🛠 **Code** | Architect → code → critic → refactor; bug hunting; reviews | `code_artisan`, `dev_squad`, `code_reviewer`, `bug_hunter` |\r\n| 🔬 **Research** | Multi-source synthesis with real citations, fact-checking | `research_lab`, `learning_tutor` |\r\n| 📊 **Data** | SQL → notebook → viz → narrative | `data_analyst` |\r\n| 🎨 **Design** | Product → UX → tech → critique | `product_studio` |\r\n| ✍️ **Writing** | Outline → draft → edit → SEO → publish | `writers_room`, `social_desk` |\r\n| 🤝 **Ops** | Triage → respond → schedule → digest | `secretary`, `concierge`, `customer_support` |\r\n| 💼 **Personal** | Calendar, finance, health, home automation | `finance`, `health_coach`, `smart_home` |\r\n| 🌐 **Social** | Outreach, support, community management | `sales_outreach`, `n8n_workflow_builder` |\r\n| 🛡 **Safety / Red-team** | Adversarial dataset generation, jailbreak research, refusal probing | *(build your own — see [data/teams/SCHEMA.md](data/teams/SCHEMA.md))* |\r\n| 🎮 **Gamify** | Agent-vs-agent, achievements, arena | *(in progress — Q4 2026)* |\r\n\r\n[Browse the 18 starter teams →](data/teams/) · [Build your own →](CONTRIBUTING.md)\r\n\r\n## Build your own team — 5-minute walkthrough\r\n\r\n1. Open **Studio** in the desktop app\r\n2. Drop in agents: orchestrator + 1..N specialists (coder, critic, researcher, brainstormer, devops, documentation, operator, …)\r\n3. Wire the dispatch graph (orchestrator → coder → critic → back to orchestrator)\r\n4. Write each agent's system prompt\r\n5. Save → team appears in your picker\r\n6. **Publish to the community** via PR against [`data/teams/`](data/teams/) — your team becomes one-click installable for every other user\r\n\r\n## Power tools nobody else ships\r\n\r\n### Fine-tune any open-weight model\r\nLoRA pipeline with Unsloth, TRL, PEFT, bitsandbytes. Llama / Qwen / Mistral / Gemma — anything on HuggingFace. Live loss curves, graceful Stop preserves checkpoints, resume-from-checkpoint and resume-adapter both supported. Runs on a 12 GB GPU.\r\n\r\n### Abliterate (refusal removal for safety research)\r\nOrthogonalise weight matrices against refusal directions — now with **effect-based (causal) selection**: it *measures* refusal on a held-out set, tests candidate directions by actually ablating and re-scoring, keeps the one that drops refusal most, and stops before over-ablating. You get a real before/after compliance number, and it works on strong RLHF models that defeat the classic single-direction recipe. Use cases:\r\n- AI safety labs training refusal classifiers need cleanly-uncensored teacher models\r\n- Red teams need models that don't sandbag jailbreak tests\r\n- Academic research on alignment failure modes\r\n\r\nThe corpus prep + abliteration script ship together.\r\n\r\n### GGUF creation + quantization\r\nConvert HF safetensors → GGUF, quantize to Q4_K_M / Q5_K_M / Q6_K / Q8_0 / F16. The same pipeline that gives you tiny, fast custom models others can run on llama.cpp / Ollama / LM Studio.\r\n\r\n### Adversarial dataset generation\r\nBuild a team whose role is to PROBE another model. Output: a labelled dataset of jailbreak attempts, refusal patterns, edge cases. Sells to AI safety labs. Trains your own filters.\r\n\r\n## Cloud or local — same teams, your choice\r\n\r\nYou don't need a 4090. Many users will never have one.\r\n\r\n- **Cloud-only:** Plug in Claude / GPT / Gemini / Kimi API keys. Teams work identically. ~30 MB install, runs on any laptop.\r\n- **Local + cloud mix:** Have a 3060? Run Llama for the bulk, hand off to Claude for the hard parts in the same conversation. Save 90% on tokens.\r\n- **Local-only:** Have a 4090? Never touch a cloud API. Privacy by default. Stop paying per token forever.\r\n\r\nSame teams. Same agent definitions. Same UI. The model layer is just plumbing.\r\n\r\n## Run anywhere\r\n\r\n| Mode | Status | Use case |\r\n|---|:---:|---|\r\n| **Desktop (Windows)** | ✅ shipped | Daily-driver AI workstation on your laptop |\r\n| **Desktop (macOS / Linux)** | 🔜 Q3 2026 | Mac / Ubuntu users |\r\n| **Headless on VPS (24/7)** | 🔜 Q4 2026 | Run your custom teams on a $5/mo box. Reach them via Telegram, web, API. Always-on agentic services. |\r\n| **Containerised / VM** | 🔜 Q4 2026 | Drop OwLLM into your existing infra. |\r\n\r\nThe team definitions, role prompts, MCP configs, and model selections are all portable across deployment modes — build a team once, run it anywhere.\r\n\r\n## Install (Windows only — for now)\r\n\r\n1. **[Download `OwLLM.Desktop.Setup.exe`](https://github.com/OwLLM/owllm/releases/latest)** (~30 MB — one file, that's it)\r\n2. Run `OwLLM-Desktop-Setup-x64.exe`. Windows SmartScreen may flag it the first time (the binary isn't EV-signed yet) — click \"More info\" → \"Run anyway\".\r\n3. On first launch, a **hardware-aware wizard** opens. It detects your hardware and offers the modules that fit:\r\n   - **Local Inference** (~33 MB CPU / ~32 MB Vulkan / ~285 MB CUDA) — only needed if you want local models\r\n   - **Audio / Speech-to-Text** (~148 MB) — for voice messages, mic input\r\n   - **Fine-tuning** (~12 GB) — only if you'll train models\r\n   - **MCP toolchain** (~260 MB) — only if you want browser / git / postgres MCP servers\r\n\r\n**Cloud-only?** Skip the wizard entirely and just enter your API keys in Settings. The shell alone is enough for cloud-model chat + agent orchestration.\r\n\r\n## How updates work\r\n\r\nThree independent update streams — small, fast, no full reinstalls:\r\n\r\n- **Shell** auto-updates via Tauri's signed updater\r\n- **Modules** (llama backend, fine-tune env, audio, MCP) check + swap per-launch\r\n- **Data layer** (team templates, role prompts, model profiles, MCP recommendations) hot-pulls from `data/` in this repo on launch. **A new team you contribute today reaches every installed app within minutes — no rebuild.**\r\n\r\nThat's why the data/ tree is open and community-driven even though the app binaries are closed-source.\r\n\r\n## ✨ Recent highlights\r\n\r\nOwLLM ships fast. Here's what landed across the **0.6.37 → 0.6.53** releases.\r\n\r\n### 🧠 Agentic memory that actually persists\r\n- **Per-agent memory** — every specialist now remembers its own prior turns across dispatches *and* across runs. Model-agnostic (local, Claude, Codex, Gemini, Kimi).\r\n- **Shared team memory (RAG)** — a durable, project-wide knowledge base the whole team pools what it learns into. **Works on every model path now** — a recent-memory snapshot is injected into every agent's prompt, and any agent (cloud CLI or API, not just local) can write a fact with a `[REMEMBER …]` line. **Syncs across your PCs.**\r\n- **3D knowledge graph** — explore that memory as a rotatable force-directed graph: 🧠 *Memory → 🌐 Graph*.\r\n\r\n### 📋 A team operating standard\r\n- Projects ship with a **native best-practice rule set** (fully editable) — enforced for every agent **and** the Critical Thinker.\r\n- A standing **operating contract**: a conflict-resolution priority order, a Definition of Done, and an agent handoff format.\r\n- Your rules **sync across machines**.\r\n\r\n### 🎓 Skills, by role, out of the box\r\n- A curated set of **skill packs auto-equipped by role** — coder, critic, researcher, writer, orchestrator… Add, remove, or install more (Anthropic / community) per agent.\r\n\r\n### ⚡ Local-model performance\r\n- **VRAM-aware context** — the context window is sized to fit your GPU *after* the model weights, so a big model stops silently spilling onto the CPU. Real, measurable speed-ups.\r\n- **Direct-drive file tools** — agents read your project folder straight off the disk instead of a slow WSL round-trip.\r\n\r\n### ⏱️ Visibility \u0026 reliability\r\n- **Run timers** — a team stopwatch in the canvas header, plus per-agent working-time on every card.\r\n- **No more idle CPU burn / freeze** — the Agents views' animations now run only while agents are working *and* the window is visible, instead of re-rendering forever (which used to peg a core and leak memory until the app locked up).\r\n- Hardened the subscription path (transient **401 / 529** auto-retry), the WSL sandbox credentials + folder handling, and post-reboot cold-start hangs.\r\n\r\n### 💬 Chat \u0026 UI polish\r\n- **Server page tells you what's live** — *Select Model* pre-selects the running model, and the context window in effect shows as a live **● Live: N tokens** badge (read from llama-server) so you never guess what's loaded.\r\n- **Cleaner team diagram** — ring distances are fixed per layer instead of stretched to fill the canvas; a single-layer team no longer flings its ring out to the outer-layer distance.\r\n- **Ctrl+A behaves** — select-all inside a chat now grabs *only* that conversation, not the whole app window. Clean copies again.\r\n- **Live tok/s in the header** — your local generation speed sits next to the API key and updates even through thinking- and tool-heavy turns.\r\n- **No more doubled output** — fixed a race that could run a team's orchestrator twice at once, interleaving two streams into one garbled reply.\r\n\r\n\u003cdetails\u003e\r\n\u003csummary\u003e\u003cb\u003eFull changelog (0.6.37 → 0.6.53)\u003c/b\u003e\u003c/summary\u003e\r\n\r\n| Version | Highlight |\r\n|---|---|\r\n| **0.6.53** | Team memory on every model path ([REMEMBER] writes + injected snapshot) · fixed runaway-CPU freeze |\r\n| **0.6.52** | Server page shows the running model + live context · fixed diagram ring distances |\r\n| **0.6.51** | Scoped Ctrl+A in chats · live tok/s in the header · no doubled orchestrator |\r\n| **0.6.50** | Effect-based abliteration — causal selection + refusal benchmark |\r\n| **0.6.49** | Context capped to fit VRAM (stops CPU spill) |\r\n| **0.6.48** | Auto-context sized from free VRAM after the model weights |\r\n| **0.6.47** | Time-bounded WSL warm-up — no post-reboot hang |\r\n| **0.6.46** | Direct-drive file tools + run timers + 529 retry |\r\n| **0.6.45** | WSL pre-flight folder-reachability check |\r\n| **0.6.44** | Agents told their project root explicitly |\r\n| **0.6.43** | 3D memory graph + sandbox-credential 401 fix |\r\n| **0.6.42** | Curated skill packs, auto-equipped by role |\r\n| **0.6.41** | Project rules sync across PCs |\r\n| **0.6.40** | Full Agent Team Rules + operating contract |\r\n| **0.6.39** | Native best-practice rule set |\r\n| **0.6.38** | Team-memory viewer + cross-PC sync |\r\n| **0.6.37** | Per-agent memory + shared RAG team memory |\r\n\r\n\u003c/details\u003e\r\n\r\n---\r\n\r\n## Roadmap\r\n\r\n- [x] Multi-agent dispatch with worktree isolation\r\n- [x] Modular installer + hardware-aware wizard\r\n- [x] MCP-first tool architecture\r\n- [x] Fine-tuning + abliteration pipeline\r\n- [x] GGUF / quantization pipeline\r\n- [x] Telegram bridge\r\n- [x] **WSL tool isolation** — agents run their tools inside Ubuntu, off your Windows drive\r\n- [x] **Folder-sealed isolation** — agents see **only the project folder** (bubblewrap inside WSL), on your **real Windows folder, no copy**\r\n- [x] **Sandbox disk management** — usage view, one-click cache clear, disk reclaim; auto-cleanup on project delete\r\n- [x] **Cloud CLIs inside the sandbox** — Claude/Codex/Gemini/Kimi run isolated too\r\n- [x] **Connect GitHub** — isolated agents clone private repos + push from inside the sandbox\r\n- [x] **Auto login-sync** — codex/claude/gemini/kimi + every API key mirrored into the sandbox\r\n- [x] **Convert projects** isolated↔not from the header; **Accounts tests host + sandbox**\r\n- [~] **Mac/Linux isolation (beta)** — Lima VM (macOS) + bubblewrap (Linux), same model as WSL\r\n- [ ] **Visual team builder** — Q3 2026\r\n- [ ] **macOS + Linux desktop** — Q3 2026\r\n- [ ] **24/7 headless / VPS mode** — Q4 2026\r\n- [ ] **Container / VM deployment** — Q4 2026\r\n- [ ] **Gamification** (agent-vs-agent arena, achievements) — Q4 2026 *(in progress)*\r\n- [ ] **WhatsApp bridge** — Q4 2026\r\n- [ ] **Vision models** (LLaVA / Pixtral) — Q4 2026\r\n- [ ] **Voice output (TTS)** — Q1 2027\r\n- [ ] **Public team marketplace** — Q1 2027\r\n\r\nTrack active work in [Discussions → Roadmap](https://github.com/OwLLM/owllm/discussions).\r\n\r\n## Who's this for\r\n\r\n- **Indie devs \u0026 founders** — your AI workforce, not a SaaS subscription\r\n- **AI safety researchers** — abliteration, red-team teams, adversarial dataset gen\r\n- **Model creators** — fine-tune, quantize, ship GGUFs\r\n- **Automation builders** — replace n8n / Zapier with agents that understand *meaning*\r\n- **Privacy-bound teams** — legal, medical, defence, regulated industries\r\n- **Agencies** — run custom client agent teams 24/7 (when VPS mode lands)\r\n- **Power users** — anyone tired of generic chatboxes\r\n\r\n## Community\r\n\r\n- 💬 [GitHub Discussions](https://github.com/OwLLM/owllm/discussions) — Q\u0026A, show what you built, roadmap input\r\n- 🐛 [Issues](https://github.com/OwLLM/owllm/issues) — bug reports (use the template)\r\n- 🎨 [Contributing](CONTRIBUTING.md) — agent teams, roles, translations, docs\r\n\r\n## License\r\n\r\nRepository contents (agent teams, role definitions, registry, schemas, docs): [MIT](LICENSE) — fork freely, share team packs, build on it.\r\n\r\nApplication binaries via [Releases](https://github.com/OwLLM/owllm/releases): see [EULA.md](EULA.md). Source for the application itself is not currently public.\r\n\r\n## Acknowledgements\r\n\r\nStanding on the shoulders of: [llama.cpp](https://github.com/ggml-org/llama.cpp), [whisper.cpp](https://github.com/ggerganov/whisper.cpp), [Tauri](https://tauri.app/), [Unsloth](https://github.com/unslothai/unsloth), [Model Context Protocol](https://modelcontextprotocol.io/), and the open-weight model creators (Meta, Alibaba, Mistral, Google, DeepSeek, Anthropic for their safety research).\r\n\r\n**If you build something cool with OwLLM, share it in [Discussions → Show \u0026 Tell](https://github.com/OwLLM/owllm/discussions/categories/show-and-tell).** Stars are how this category proves itself worth investing in.\r\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fowllm%2Fowllm","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fowllm%2Fowllm","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fowllm%2Fowllm/lists"}