Tilde.run: An Agent Sandbox with a Transactional, Versioned Filesystem

Tilde.run is a sandbox for AI agents that turns every agent run into a reversible transaction. It creates a single versioned filesystem from multiple data sources — GitHub, S3, and Google Drive are mounted as a single ~/sandbox directory. Every file is versioned from the first commit, and any agent run can be rolled back instantly with one command.
Key Features
Versioned Composable Filesystem
A real POSIX filesystem—any tool, any language, no SDKs. Mount code from GitHub, training data from S3, and documents from Google Drive as a single ~/sandbox. Example mounts:
4 mounts
~/sandbox
├── code github acme/ml-pipeline
├── data s3 acme-data/training
├── docs gdrive team-wiki
└── output local
All versioned, all reversible.
Safe Serverless Sandboxes
Each run is a transaction in a fresh, isolated container. On clean exit, changes commit atomically; on failure, nothing changes. No backups to restore, no manual cleanup.
Network Isolation
Cloud metadata, private networks, and unauthorized hosts are blocked by default. Every outbound request is policy-checked and logged. Example egress log:
12:04:01 GET api.openai.com /v1/completions ALLOW
12:04:03 POST api.anthropic.com /v1/messages ALLOW
12:04:05 GET pypi.org /simple/pandas ALLOW
12:04:07 POST evil-exfil.io /upload DENY
12:04:08 GET 169.254.169.254 /metadata DENY
12:04:09 PUT registry.npmjs.org /my-pkg DENY
Time Travel & Audit Trail
Browse the full timeline, inspect diffs, and revert any commit instantly. Every change is tied to the human, process, or agent that produced it.
Agent-first RBAC
Agents have scoped permissions—never full user access. Granular policies in a simple DSL. Example policy:
analyst-policy
GetObject (path: "/data/*") ?
PutObject (path: "/reports/*") # require human approval!
! PutObject (path: "/secrets/*")
Quickstart
Install in one line:
$ curl -fsSL https://tilde.run/install | sh
Run an agent in a sandbox:
$ tilde exec my-team/documents \
--image python:3.12 \
-- /sandbox/code/agent.py --input /sandbox/data/reports
sandbox running...
sandbox completed. exit code: 0, commit id: c9d0e1f2
Or start an interactive shell:
$ tilde shell my-team/documents --image python:3.12
root@sb-7f3a9c01:/sandbox$ _
Python SDK:
import tilde
repo = tilde.repository("my-team/documents")
with repo.shell(image="python:3.12") as sh:
sh.run("pip install pandas")
result = sh.run("python agent.py --input /sandbox/data")
print(result.stdout.text())
Integrates with Claude, AWS S3, LangGraph, Google Drive, and Hugging Face.
📖 Read the full source: HN AI Agents
👀 See Also

ClankerRank: A Benchmark for AI-Assisted Coding Skills with Claude Haiku
A developer built ClankerRank to measure proficiency in AI-assisted coding using Claude's Haiku 4.5 model. The platform presents identical bugs to users, scores outputs with hidden test suites, and has revealed clear skill gaps among hundreds of participants.

Mandala v0.3: Open-Source Async Runtime to Unify Logistics Telemetry as OpenTelemetry Spans for Agent Reasoning
Mandala v0.3 provides an open-source async runtime that ingests telemetry from Samsara, Descartes, Vizion, and FMCSA via webhooks, emits events as OpenTelemetry spans, and exposes data via MCP tools for LLM agents.

Multi-Agent Debate Approach Improves LLM Reasoning Quality
A developer experimented with a multi-agent debate approach using CyrcloAI, where different AI agents take on roles like analyst, critic, and synthesizer to critique each other's responses before producing a final answer, resulting in more structured and deliberate outputs.

CopilotKit: Open-Source React Building Blocks for Agent UIs
CopilotKit (30k stars, MIT) provides React components for agent UI layer: chat, streaming, tool calls, human-in-the-loop, and generative UI, with AG-UI protocol support across LangGraph, ADK, CrewAI, and more.