@wenathlan/saddle 1.8.17 → 2.0.2

package/docs/talks1/assistant-05.md

@@ -1,774 +0,0 @@
-# assistant
-
-[Thinking] The user wants me to continue researching and create a comprehensive todo list of all the topics mentioned. They want me to research:
-
-Git hosting platforms: GitHub, GitLab, Forgejo, Gitea, Codeberg
-ML/AI platforms: Hugging Face, Kaggle, ModelScope
-Databases: Drizzle ORM, Prisma, MySQL2
-Node.js, Python, Rust frameworks
-Memory emulation/bridging (storage to RAM conversion) - zram, tmpfs, frontswap, zswap
-GPU emulation - virtio-gpu, virglrenderer, rutabaga_gfx
-Docker sandboxes
-Pipelines, cron jobs
-Open source projects that do this alreadyLet me continue with more searches to be comprehensive.
-
-[Thinking] Now let me search for more specific topics the user mentioned - npm packages, Python libraries, Rust tools, etc.
-
-[Thinking] Now I have comprehensive research on all the topics. Let me also search for a few more specific topics the user mentioned:
-
-Archbox and other sandbox alternatives
-Open source projects that combine all these
-npm packages for memory management
-
-[Thinking] Now I have comprehensive research on all the topics. Let me compile a comprehensive todo list for the user.
-
-Continuando a pesquisa completa. Vou buscar cada tópico mencionado:
-
-Aqui está a TODO LIST COMPLETA de pesquisa, organizada por categorias:
-
-TODO LIST DE PESQUISA COMPLETA
-1. GIT HOSTING + CI/CD (Infraestrutura Gratuita)
-
-
-
-#
-Plataforma
-RAM Gratuita
-CPU
-Disk
-Tempo
-URL
-
-
-
-1.1
-GitHub Actions
-7GB
-2 cores
-14GB
-6h/job
-github.com/features/actions
-
-
-1.2
-GitLab CI
-8GB
-2 cores
-30GB
-50min/job
-about.gitlab.com/pricing
-
-
-1.3
-Gitea Actions (self-hosted)
-Ilimitado*
-Ilimitado*
-Ilimitado*
-Sem limite
-gitea.com
-
-
-1.4
-Forgejo Actions (self-hosted)
-Ilimitado*
-Ilimitado*
-Ilimitado*
-Sem limite
-forgejo.org
-
-
-1.5
-Codeberg Woodpecker CI
-2-4GB
-2-4 cores
--
-2-4min
-codeberg.org
-
-
-1.6
-Codeberg Forgejo Actions
-2-4GB
-4 cores
-4GB
-4min
-ci.codeberg.org
-
-
-*Depende do hardware do host
-2. ML/AI PLATFORMS (Compute Gratuito)
-
-
-
-#
-Plataforma
-RAM Gratuita
-CPU/GPU
-Disk
-Tempo
-URL
-
-
-
-2.1
-Hugging Face Spaces
-16GB
-2 cores
-50GB
-Sem limite*
-huggingface.co/docs/hub
-
-
-2.2
-Kaggle Notebooks
-29-30GB
-4 cores + P100/T4 GPU
-20GB
-12h/session, 30h/semana GPU
-kaggle.com
-
-
-2.3
-ModelScope
-Variável
-CPU/GPU
-Variável
-Variável
-modelscope.cn
-
-
-2.4
-HF ZeroGPU
-16GB
-H200 slices
-50GB
-3.5min/dia (free)
-huggingface.co
-
-
-*HF Free dorme após 48h de inatividade
-3. SANDBOX/CONTAINER OPEN SOURCE (Projetos)
-
-
-
-#
-Projeto
-Linguagem
-Tipo
-RAM
-Boot Time
-URL
-
-
-
-3.1
-CubeSandbox (TencentCloud)
-Rust+Go
-KVM MicroVM
-<5MB overhead
-<60ms
-github.com/TencentCloud/CubeSandbox
-
-
-3.2
-Firecracker (AWS)
-Rust
-KVM MicroVM
-<5MB
-<150ms
-github.com/firecracker-microvm/firecracker
-
-
-3.3
-E2B
-TypeScript
-Cloud sandbox
--
-<100ms
-e2b.dev
-
-
-3.4
-sandboxd
-Go
-Docker sandbox
--
-~1s
-github.com/tastyeffectco/sandboxd
-
-
-3.5
-Containarium
-Go
-LXC containers
-Variável
-~1s
-github.com/footprintai/containarium
-
-
-3.6
-HiveBox
-Rust
-Linux namespaces
-2-5MB
-10-50ms
-github.com/TetiAI/hivebox
-
-
-3.7
-Nucleus
-Rust
-OCI runtime
-Mínimo
-Rápido
-github.com/sig-id/nucleus
-
-
-3.8
-Crucible
-Go
-Firecracker
-<5MB
-<170ms
-github.com/gnana997/crucible
-
-
-3.9
-Mitos
-Go+Rust
-Firecracker + K8s
-<5MB
-~27ms
-github.com/mitos-run/mitos
-
-
-3.10
-arch-sandbox
-Go
-systemd-nspawn
-Mínimo
-Rápido
-github.com/OminduD/arch-sandbox
-
-
-3.11
-gitvm
-Go
-Firecracker+Docker
-Variável
-~125ms
-github.com/open-gitagent/gitvm
-
-
-4. SANDCastle (TypeScript Sandbox Orchestrator)
-
-
-
-#
-Pacote
-Descrição
-URL
-
-
-
-4.1
-@ai-hero/sandcastle
-Orquestra coding agents em sandboxes
-npmjs.com/package/@ai-hero/sandcastle
-
-
-4.2
-sandbox-orchestrator
-API Node.js para execução Docker
-github.com/dea1j/sandbox-orchestrator
-
-
-4.3
-agentic-control
-Orquestração multi-agent com sandbox
-npmjs.com/package/@jbcom/agentic
-
-
-5. DATABASES (ORMs)
-
-
-
-#
-ORM
-Drivers Suportados
-URL
-
-
-
-5.1
-Drizzle ORM
-SQLite, PostgreSQL, MySQL, Turso, Neon, D1, etc.
-orm.drizzle.team
-
-
-5.2
-Prisma
-PostgreSQL, MySQL, SQLite, MongoDB
-prisma.io
-
-
-5.3
-mysql2
-MySQL nativo
-npmjs.com/package/mysql2
-
-
-6. MEMÓRIA: Storage → RAM Bridge
-
-
-
-#
-Tecnologia
-Descrição
-URL
-
-
-
-6.1
-zram
-Dispositivo bloco RAM-comprimido
-docs.kernel.org/admin-guide/blockdev/zram.html
-
-
-6.2
-zswap
-Cache comprimido para swap
-kernel.org/doc/html/v6.3/admin-guide/mm/zswap.html
-
-
-6.3
-tmpfs
-Sistema de arquivos em RAM
-(nativo Linux)
-
-
-6.4
-Frontswap
-Interface "transcendent memory"
-docs.kernel.org/6.0/mm/frontswap.html
-
-
-6.5
-zram-config
-Utilitário completo zram (Raspberry Pi)
-github.com/ecdye/zram-config
-
-
-6.6
-node-shared-mem
-Shared memory para Node.js
-npmjs.com/package/node-shared-mem
-
-
-6.7
-shmmap.js
-mmap + POSIX shm para Node.js
-npmjs.com/package/shmmap
-
-
-6.8
-shm-typed-array
-IPC shared memory Node.js
-npmjs.com/package/shm-typed-array
-
-
-7. GPU EMULATION / VIRTUALIZATION
-
-
-
-#
-Projeto
-Descrição
-URL
-
-
-
-7.1
-virtio-gpu (QEMU)
-GPU paravirtualizada
-qemu.readthedocs.io
-
-
-7.2
-virglrenderer
-OpenGL emulado via Gallium3D
-gitlab.freedesktop.org/virgl
-
-
-7.3
-rutabaga_gfx
-Abstração GPU cross-platform (Rust)
-github.com/magma-gpu/rutabaga_gfx
-
-
-7.4
-vhost-device-gpu
-Daemon GPU emulation (Rust)
-crates.io/crates/vhost-device-gpu
-
-
-7.5
-virtio-win drivers
-Drivers GPU para Windows guests
-github.com/arehnman/kvm-guest-drivers-windows
-
-
-8. PYTHON SANDBOX LIBRARIES
-
-
-
-#
-Pacote
-Descrição
-URL
-
-
-
-8.1
-python-sandbox (Onyx)
-Execução Python segura em Docker
-github.com/onyx-dot-app/python-sandbox
-
-
-8.2
-agt_sandbox (Microsoft)
-Sandbox multi-backend (Docker/HyperLight/ACA)
-pypi.org/project/agt_sandbox
-
-
-8.3
-onit-sandbox
-MCP server sandbox Python
-github.com/sibyl-oracles/onit-sandbox
-
-
-8.4
-mcp-sandbox
-Python sandbox para LLMs
-github.com/johanli233/mcp-sandbox
-
-
-8.5
-rlm-python
-Execução código LLM com Docker sandbox
-pypi.org/project/rlm-python
-
-
-9. RUST SANDBOX CRATES
-
-
-
-#
-Crate
-Descrição
-URL
-
-
-
-9.1
-firecracker
-SDK Rust para Firecracker API
-crates.io/crates/firecracker
-
-
-9.2
-fctools
-SDK modular Firecracker
-crates.io/crates/fctools
-
-
-9.3
-firecracker-rs-sdk
-SDK Firecracker (std/tokio/async-std)
-github.com/xuehaonan27/firecracker-rs-sdk
-
-
-10. NODE.JS FRAMEWORKS + FERRAMENTAS
-
-
-
-#
-Pacote
-Descrição
-URL
-
-
-
-10.1
-dockerode
-Cliente Docker para Node.js
-npmjs.com/package/dockerode
-
-
-10.2
-docker-compose
-Orquestração Docker
-npmjs.com/package/docker-compose
-
-
-10.3
-@huggingface/inference
-API Hugging Face
-npmjs.com/package/@huggingface/inference
-
-
-11. Z.AI / ChatGLM INFRAESTRUTURA
-
-
-
-#
-Projeto
-Descrição
-URL
-
-
-
-11.1
-GLM-5
-Modelo Z.AI com deploy Docker
-github.com/zai-org/GLM-5
-
-
-11.2
-Z.ai2api
-Proxy OpenAI-compatible
-github.com/hmjz100/Z.ai2api
-
-
-11.3
-GLM-ZAI-2API
-Proxy Go para Z.AI
-github.com/D3-vin/GLM-ZAI-2API
-
-
-11.4
-Oxide-Agent
-Bot Telegram com sandbox Rust
-github.com/0FL01/Oxide-Agent
-
-
-11.5
-sandcastle-zai
-Sandcastle com provider Z.AI
-github.com/ShavedTundra/sandcastle-zai
-
-
-12. MULTI-AGENT ORCHESTRATION
-
-
-
-#
-Projeto
-Descrição
-URL
-
-
-
-12.1
-Clawix
-Multi-agent com Docker isolamento
-github.com/ClawixAI/clawix
-
-
-12.2
-stateful-agentic-sandbox
-Node.js + Python sandbox
-github.com/Daniel5569/stateful-agentic-sandbox-orchestrator
-
-
-12.3
-E2B alternatives
-Comparativo completo
-amux.io/guides/ai-agent-sandboxing
-
-
-13. DOCKER HARDENING
-
-
-
-#
-Recurso
-Comando/Flag
-
-
-
-13.1
-Network isolation
---network=none
-
-
-13.2
-Drop capabilities
---cap-drop=ALL
-
-
-13.3
-No privilege escalation
---security-opt=no-new-privileges
-
-
-13.4
-Read-only rootfs
---read-only
-
-
-13.5
-Memory limit
---memory=8g
-
-
-13.6
-CPU limit
---cpus=4
-
-
-13.7
-PID limit
---pids-limit=512
-
-
-13.8
-gVisor runtime
---runtime=runsc
-
-
-14. PROJETOS COMPARATIVOS (quem faz o quê)
-
-
-
-Projeto
-Isolamento
-Startup
-Overhead
-Self-hosted
-Open Source
-
-
-
-Docker
-Namespaces
-<1s
-<2% CPU
-Sim
-Apache 2.0
-
-
-Firecracker
-KVM MicroVM
-<150ms
-<5MB
-Sim
-Apache 2.0
-
-
-CubeSandbox
-KVM + eBPF
-<60ms
-<5MB
-Sim
-Apache 2.0
-
-
-HiveBox
-NS + seccomp + Landlock
-10-50ms
-2-5MB
-Sim
-MIT
-
-
-Crucible
-Firecracker
-<170ms
-<5MB
-Sim
-Apache 2.0
-
-
-gVisor
-Syscall emulation
-<1s
-5-15%
-Sim
-Apache 2.0
-
-
-E2B
-Firecracker (cloud)
-<100ms
-<5MB
-Parcial
-Parcial
-
-
-15. DEPLOYMENT + ORCHESTRATION
-
-
-
-#
-Ferramenta
-Descrição
-
-
-
-15.1
-Docker Compose
-Orquestração local multi-container
-
-
-15.2
-Kubernetes
-Orquestração clusters
-
-
-15.3
-Nomad
-Orquestração HashiCorp
-
-
-15.4
-systemd-nspawn
-Containers leves Linux
-
-
-15.5
-LXC/Incus
-System containers
-
-
-
-RESUMO: ESTRATÉGIA RECOMENDADA
-Para sandbox gratuito com infraestrutura de terceiros:
-
-GitHub/GitLab Actions para builds temporários
-Hugging Face Spaces para apps web persistentes
-Kaggle para workloads pesados com GPU
-
-Para sandbox self-hosted (mais controle):
-
-CubeSandbox (melhor: <60ms boot, <5MB overhead, E2B-compatible)
-Firecracker (proven: AWS Lambda usa)
-HiveBox (mais leve: 2-5MB, 10-50ms)
-sandboxd (mais simples: Go, Docker, SQLite)
-
-Para memória virtual:
-
-zram + tmpfs (nativo Linux)
-node-shared-mem / shmmap.js (Node.js)
-
-Para banco de dados:
-
-Drizzle ORM (mais leve, TypeScript)
-Prisma (mais features)
-
-Para GPU virtual:
-
-virglrenderer (OpenGL)
-rutabaga_gfx (cross-platform)
-
-Quer que eu implemente alguma parte específica?