惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

D
Docker
Simon Willison's Weblog
Simon Willison's Weblog
H
Help Net Security
F
Fortinet All Blogs
H
Heimdal Security Blog
S
Schneier on Security
L
LangChain Blog
博客园 - Franky
酷 壳 – CoolShell
酷 壳 – CoolShell
NISL@THU
NISL@THU
P
Palo Alto Networks Blog
J
Java Code Geeks
博客园 - 【当耐特】
The Last Watchdog
The Last Watchdog
W
WeLiveSecurity
www.infosecurity-magazine.com
www.infosecurity-magazine.com
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
V
Vulnerabilities – Threatpost
I
InfoQ
Recorded Future
Recorded Future
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
C
CERT Recently Published Vulnerability Notes
T
Tenable Blog
腾讯CDC
C
Check Point Blog
量子位
M
MIT News - Artificial intelligence
GbyAI
GbyAI
罗磊的独立博客
Threat Intelligence Blog | Flashpoint
Threat Intelligence Blog | Flashpoint
B
Blog
小众软件
小众软件
cs.AI updates on arXiv.org
cs.AI updates on arXiv.org
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
C
CXSECURITY Database RSS Feed - CXSecurity.com
Stack Overflow Blog
Stack Overflow Blog
P
Proofpoint News Feed
P
Privacy & Cybersecurity Law Blog
V2EX - 技术
V2EX - 技术
T
Threatpost
Engineering at Meta
Engineering at Meta
Attack and Defense Labs
Attack and Defense Labs
T
Tailwind CSS Blog
S
Securelist
The Cloudflare Blog
博客园 - 叶小钗
L
LINUX DO - 最新话题
T
Troy Hunt's Blog
C
Cyber Attacks, Cyber Crime and Cyber Security
爱范儿
爱范儿

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor GitHub - GenAI-Gurus/awesome-eu-ai-act: Curated tools, official sources, OSS, templates, and guides for EU AI Act compliance. Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders How to Switch AI Chatbots and Why You Might Want To GitHub - MattMessinger1/agentic_refund_guardrail: Safe refund policy layer for AI agents — Python + TypeScript. Same behavior, shared tests. Adam/papers/emergent_values_whitepaper.md at master · strangeadvancedmarketing/Adam Ask HN: How do you stop playing 20 questions with your AI coding tools How far can automation and AI support psychotherapy? - @theU GitHub - stagas/rtdiff: realtime git diff gui and AI-assisted commits A Mac Studio for Local AI — 6 Months Later A History of the Early Years of AI at the University of Edinburgh Why AI Coding Tools Still Feel Stuck on Localhost MSN AI Datacenters Are Becoming Strategic Targets twitter.com Penn Researchers Use AI to Surface Unreported GLP-1 Side Effects in Reddit Posts Show HN: MoodSense AI (ML and FastAPI and Gradio, Deployed on Hugging Face) Moodsense Ai - a Hugging Face Space by aman179102 AI models are terrible at betting on soccer—especially xAI Grok GitHub - xialeistudio/echoic GitHub - HimashaHerath/github-dev-wrapped: AI-powered weekly GitHub activity reports deployed to GitHub Pages GitHub - alejandrobalderas/claude-code-from-source: Architecture, patterns & internals of Anthropic's AI coding agent — reverse-engineered from source maps AI and Tech brief: Ireland ascendant GitHub - Titovilal/context0: Context0 - Never Surrender Training for a Marathon with an AI Coach: What Worked and What Didn't Cyber Pulse: Agentic Intel - Apps on Google Play I Built an AI PR Reviewer That Catches Bugs by Not Looking for Bugs Gen Z workers are so fearful AI will take their job they’re intentionally sabotaging their company’s AI rollout | Fortune How AI Is Reimagining the Game of Golf–For Both Players and Courses GitHub - nattergabriel/reseed: A CLI tool for managing and distributing agent skills across projects Is SVG the final frontier? My AI workflow evolved from prompts to a near-autonomous workflow MLSharp Help - 3DGS Viewer & Generator I put my cognitive field based AI's runtime on GitHub Is Numble the first AI-proof game? A3: Kubernetes for autonomous AI agent fleets | Emergent Principles Deepali Vyas ("The Elite Recruiter") GitHub - msmarkgu/RelayFreeLLM: A restful API designed to route user prompts to various AI model providers. Unionized ProPublica staff are on strike over AI, layoffs, and wages Unleashing the Advantage of Quantum AI We're heading for an AI-fueled 'dementia crisis,' brain scientist warns The AI-Assisted Breach of Mexico's Government Infrastructure [pdf] GitHub - stef41/lmscan: 🔍 Detect AI-generated text and fingerprint which LLM wrote it. Open-source GPTZero alternative. Zero dependencies, works offline. MSN GitHub - visionscaper/collabmem: Enabling long-term collaboration with Agentic AI - building up episodic and world model memory over time with in-context awareness We gave an AI a 3 year retail lease in SF and asked it to make a profit | Andon Labs AI Code is Hollowing Out Open Source, and Maintainers are Looking the Other Way What leaked "SteamGPT" files could mean for the PC gaming platform's use of AI AI is the boss at this retail store. What could go wrong? GitHub - Wuzu11517/agentic-proxy: Local proxy meant to help reduce With Drones, Geophysics and ArtificiaI Intelligence, Researchers Prepare to Do Battle Against Land Mines A Single Operator, Two AI Platforms, Nine Government Agencies: The Full Technical Report 在 Steam 上购买 FriedrichAI: Offline AI 立省 10% GitHub - inevolin/resume-cli: Hit Claude usage limits? Resume any AI coding session elsewhere. Switch tools at zero friction. GitHub - atripati/ark: AI Runtime Kernel — a context operating system for AI agents. Eliminates tool bloat, loads only what’s needed, and gives LLMs their reasoning space back. How to Build a Secure AI PR Reviewer with Claude, GitHub Actions, and JavaScript This Startup Wants You to Pay Up to Talk With AI Versions of Human Experts Intel Arc Pro B70 Brings 32GB VRAM to Local AI for $949 WordPress 7.0: The Good, the AI, and the Still Missing AI on the couch: Anthropic gives Claude 20 hours of psychiatry IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures AI Agents Know About Supabase. They Don't Always Use It Right. The history and future of AI at Google, with Sundar Pichai Inside an AI‑enabled device code phishing campaign How Meta Used AI to Map Tribal Knowledge in Large-Scale Data Pipelines AI for Systems: Using LLMs to Optimize Database Query Execution Forecasting the Economic Effects of AI Introducing Tinker: Play with AI, bring your ideas to life AI sheds light on an ancient gaming mystery People really hate AI but not as much as Iran—or Democrats | Fortune What is an AI Product Engineer? Phoebe Gates wants her $185 million AI startup to succeed with 'no ties to my privilege or my last name': 'I have a chip on my shoulder' | Fortune
GitHub - CelestoAI/smolfs: Durable workspace folders for AI agents.
theaniketmau · 2026-06-25 · via Hacker News - Newest: "AI"

Durable workspace folders for AI agents.

CI Publish CLI Publish Python Package License Rust 1.85+ Python 3.9+

Quickstart | Lifecycle | Python SDK | TypeScript SDK | Development | Releases


SmolFS gives agents a workspace folder that can survive after the agent process stops. You mount it like a normal directory, write files into it, unmount it when the job is done, and mount it again later from another process.

SmolFS owns the developer experience around creating volumes, checking the machine, mounting, flushing, unmounting, and inspecting status. The storage backend is installed and managed for you.


Durable workspaces

Agents can keep files across short-lived runtimes without each runtime managing storage setup directly.

Read more ->

Local dev mode

Use --dev for a local-only volume backed by SQLite metadata and local object files.

Read more ->

Cloud volumes

Use Redis plus S3-compatible object storage when the same workspace needs to outlive one machine.

Read more ->

One CLI lifecycle

Run doctor, init, mount, flush, status, and unmount from one command.

Read more ->

Thin SDKs

Python and TypeScript bindings call the same Rust core so agent tools can use SmolFS without shelling out.

Read more ->

Explicit configuration

Cloud metadata, buckets, and credentials stay explicit so durable agent data is easier to audit.

Read more ->

Use Cases

  • Keep agent work across turns. Mount the same workspace again after an agent process exits.
  • Share a workspace across runtimes. Put metadata in Redis and file contents in S3-compatible storage.
  • Test locally before using cloud storage. Start with --dev, then switch to explicit metadata and object storage settings.
  • Wrap storage in agent tooling. Use the Python or TypeScript SDK from an agent runner instead of teaching every agent process about storage internals.

Quickstart

Install SmolFS:

curl -fsSL https://raw.githubusercontent.com/CelestoAI/smolfs/main/scripts/install.sh | sh

The installer downloads the latest CLI release asset for your platform and installs SmolFS' managed storage backend. If no release asset exists yet, use the source checkout flow in Development. Set SMOLFS_INSTALL_BACKEND=0 if you only want to install the CLI. Set SMOLFS_VERSION=dev to try the latest successful build from main; tagged v* releases remain the stable channel.

Check the machine and try a local volume:

smolfs doctor
smolfs init demo --dev
smolfs mount demo ./workspace
echo hello > ./workspace/hello.txt
smolfs flush demo
smolfs unmount demo
smolfs mount demo ./workspace
cat ./workspace/hello.txt

SmolFS needs local mount support on the machine that mounts volumes. Mount support lets SmolFS provide a folder your tools can read and write.

Install the Python SDK with the CLI
curl -fsSL https://raw.githubusercontent.com/CelestoAI/smolfs/main/scripts/install.sh | SMOLFS_INSTALL_PYTHON=1 sh

The installer runs uv add smolfs from a directory with pyproject.toml, or uv pip install smolfs inside an active virtualenv. Set SMOLFS_PYTHON_MODE=user to use uv pip install --user smolfs.

CLI Lifecycle

The normal SmolFS flow is: check the machine, create a volume, mount it, use the directory, flush important writes, unmount it, and mount it again when you need the same files later.

Check the Machine

Run doctor before creating or mounting volumes:

doctor checks whether SmolFS has its managed storage backend and whether the machine can mount local directories.

Useful options:

  • smolfs doctor --install installs SmolFS' managed storage backend.
  • smolfs doctor --json prints the same report as JSON for scripts.

SmolFS looks for its home directory in SMOLFS_HOME. If it is not set, SmolFS uses ~/.smolfs. The home directory stores SmolFS config, volume records, logs, managed binaries, and local dev-volume data.

Create a Local Volume

A volume is the named workspace that SmolFS can mount later.

--dev creates a local-only volume. It uses SQLite for metadata and local files for object data under the SmolFS home directory. This is the simplest path for trying SmolFS on one machine.

Create a Cloud Volume

Cloud volumes need explicit metadata and object storage settings. Metadata stores the file tree. Object storage is where file contents live.

smolfs init agent-workspace \
  --metadata redis://localhost:6379/1 \
  --storage s3 \
  --bucket https://my-bucket.s3.us-east-2.amazonaws.com

You can pass object storage in either of these forms:

  • --store s3://bucket/prefix, --store gs://bucket/prefix, or --store file:///path/to/objects.
  • --storage TYPE --bucket BUCKET, which is useful for S3-compatible services that expect an endpoint-style bucket URL.

For Cloudflare R2 or another S3-compatible service, keep credentials in the environment used by SmolFS. Do not put access keys in command arguments or logs.

set -a
. ./.env
set +a

export SMOLFS_HOME=/tmp/smolfs-r2-home
VOL="r2demo-$(date +%Y%m%d%H%M%S)"

smolfs init "$VOL" \
  --metadata "$SMOLFS_R2_METADATA" \
  --storage s3 \
  --bucket "$SMOLFS_R2_BUCKET_URL"

Mount and Use the Volume

Mounting makes the volume appear as a normal local directory:

smolfs mount demo ./workspace
echo hello > ./workspace/hello.txt
cat ./workspace/hello.txt

SmolFS creates the mount directory if it does not exist. After the mount succeeds, programs can read and write files through that directory.

Useful options:

  • --check-storage asks SmolFS to test object storage access before the mount completes.
  • --foreground runs the mount process in the foreground instead of starting it in the background.

Flush, Inspect, and Unmount

Run flush when you want a best-effort check that recent writes have reached the mounted filesystem:

Run status to see known volumes:

smolfs status
smolfs status demo
smolfs status --json

Unmount when the job is done:

unmount asks SmolFS to flush before detaching the mountpoint. Use smolfs umount demo if you prefer the shorter alias. Add --force when the mountpoint is busy and you want SmolFS to force the unmount.

After unmounting, you can mount the same volume again and read the files:

smolfs mount demo ./workspace
cat ./workspace/hello.txt

Commands

Command What it does
smolfs doctor Checks SmolFS storage, local mount support, and configuration.
smolfs init NAME --dev Creates a local development volume.
smolfs init NAME --metadata URL --storage TYPE --bucket BUCKET Creates a cloud volume with explicit metadata and object storage.
smolfs mount NAME PATH Mounts a volume at a local directory.
smolfs flush NAME Probes the mounted volume and syncs a small file through it.
smolfs status [NAME] Shows known volumes and current mountpoints.
smolfs unmount NAME Unmounts a mounted volume and asks SmolFS to flush.
smolfs umount NAME Alias for smolfs unmount NAME.

Every command has its own help page:

smolfs help
smolfs init --help

Python SDK

The Python package is SDK-only. Install it with uv:

For local development from this checkout:

uv run --isolated --with-editable ./bindings/python python -c "from smolfs import doctor; print(doctor())"

Use the SDK from any Python agent runner:

from pathlib import Path

from smolfs import SmolFS, doctor

report = doctor()
if not report["storage_backend"]["found"] or not report["mount_support"]["found"]:
    raise RuntimeError(f"SmolFS is not ready: {report}")

fs = SmolFS.from_env()
volume = fs.ensure_volume("demo", dev=True)
mount = fs.mount(volume.name, "./workspace")

workspace = Path(mount.mountpoint)
(workspace / "hello.txt").write_text("hello from SmolFS\n")

try:
    fs.flush(volume.name)
finally:
    fs.unmount(volume.name)

Cloud volumes use the same API with explicit metadata and object storage:

fs.ensure_volume(
    "agent-workspace",
    metadata="redis://localhost:6379/1",
    storage="s3",
    bucket="https://my-bucket.s3.us-east-2.amazonaws.com",
)

For S3-compatible services such as MinIO or Cloudflare R2, pass the service bucket URL and provide ACCESS_KEY and SECRET_KEY in the environment used by SmolFS.

TypeScript SDK

The TypeScript package is a native Node.js binding over the same Rust core. The npm package ships prebuilt native bindings for Linux and macOS on x86_64 and arm64.

Install it with npm:

npm install @celestoai/smolfs

For local development from this checkout, use Node.js 18 or newer:

cd bindings/node
npm ci
npm run build:debug
npm test

Use the SDK from a Node.js agent runner:

import { SmolFS, doctor } from "@celestoai/smolfs";
import { writeFile } from "node:fs/promises";
import { join } from "node:path";

const report = doctor();
if (!report.storageBackend.found || !report.mountSupport.found) {
  throw new Error(`SmolFS is not ready: ${JSON.stringify(report)}`);
}

const fs = SmolFS.fromEnv();
const volume = fs.ensureVolume({ name: "demo", dev: true });
const mount = fs.mount({ name: volume.name, path: "./workspace" });

try {
  await writeFile(join(mount.mountpoint, "hello.txt"), "hello from SmolFS\n");
  fs.flush(volume.name);
} finally {
  fs.unmount(volume.name);
}

Cloud volumes use the same options object:

fs.ensureVolume({
  name: "agent-workspace",
  metadata: "redis://localhost:6379/1",
  storage: "s3",
  bucket: "https://my-bucket.s3.us-east-2.amazonaws.com"
});

Development

Work from a source checkout when you are changing SmolFS itself or when a CLI release asset has not been published yet.

Build and check the CLI:

cargo build -p smolfs-cli
./target/debug/smolfs doctor

Run the normal quality checks:

cargo fmt --all -- --check
cargo clippy --workspace -- -D warnings
cargo test --workspace

Run the R2-style lifecycle from this checkout:

cargo build -p smolfs-cli

set -a
. ./.env
set +a

export SMOLFS_HOME=/tmp/smolfs-r2-home
VOL="r2demo-$(date +%Y%m%d%H%M%S)"
MOUNT="/tmp/smolfs-r2-workspace"

./target/debug/smolfs init "$VOL" \
  --metadata "$SMOLFS_R2_METADATA" \
  --storage s3 \
  --bucket "$SMOLFS_R2_BUCKET_URL"

./target/debug/smolfs mount "$VOL" "$MOUNT"
echo "hello from smolfs r2" > "$MOUNT/hello.txt"
./target/debug/smolfs flush "$VOL"
./target/debug/smolfs unmount "$VOL"
./target/debug/smolfs mount "$VOL" "$MOUNT"
cat "$MOUNT/hello.txt"

Run the MinIO integration test path when the SmolFS storage backend, Redis, and a MinIO bucket are available:

SMOLFS_RUN_INTEGRATION=1 cargo test --workspace -- --nocapture

Build the Python wheel:

uvx maturin build --manifest-path bindings/python/Cargo.toml --interpreter python

Develop the Python binding locally:

uvx maturin develop --manifest-path bindings/python/Cargo.toml

Test the TypeScript SDK:

cd bindings/node
npm ci
npm test

Security and Reliability

SmolFS stores agent workspace data outside the sandbox lifecycle. Treat it like durable infrastructure.

  • Do not log credentials, S3 access keys, Redis URLs with secrets, or mount tokens.
  • Prefer explicit object-store configuration over hidden global state.
  • Make mount and unmount behavior idempotent where possible.
  • Fail loudly on missing storage backend, missing metadata URLs, missing object-store config, or missing local mount support.
  • Avoid changes that weaken persistence guarantees without calling them out.

Releases

The smolfs command is built from the Rust CLI crate. The GitHub workflow at .github/workflows/publish-cli.yml builds Linux and macOS release binaries for x86_64 and arm64 targets, smoke-tests smolfs --help, and attaches tarballs to v* GitHub releases. Pushes to main publish a rolling dev prerelease after deleting the previous dev release and tag, so the dev channel only exposes the latest successful main build.

Python packaging uses uv and maturin. The GitHub workflow at .github/workflows/publish-python.yml builds wheels for Linux and macOS, builds an sdist, and publishes to PyPI when a v* tag is pushed.

Node.js packaging uses npm and napi-rs. The GitHub workflow at .github/workflows/publish-node.yml builds native TypeScript SDK bindings for Linux and macOS, checks the npm package contents, and publishes to npm when a v* tag is pushed.

Before the first Python release, configure PyPI Trusted Publishing:

  1. Create or claim the smolfs project on PyPI.
  2. Add a trusted publisher for repository CelestoAI/smolfs.
  3. Set the workflow name to publish-python.yml.
  4. Set the environment name to pypi.

Before the first npm release, configure npm trusted publishing for @celestoai/smolfs:

  1. Add a trusted publisher for GitHub Actions.
  2. Set the organization or user to CelestoAI.
  3. Set the repository to smolfs.
  4. Set the workflow filename to publish-node.yml.
  5. Set the environment name to npm.
  6. Allow npm publish.

Release:

git tag v0.1.0
git push origin v0.1.0

Roadmap

  • Add type stubs for the Python package.
  • Add a Linux CI job that mounts a local dev volume when mount support is available.
  • Add release notes and a changelog before the first non-draft release.

License

Apache 2.0.