Skip to main content
Live
Main content

DigitalOcean launches Cuttlefish, an iPhone app for supervising AI agents

The app streams live agent sessions and lets developers approve tool calls from anywhere, while Box Mount syncs governed Box folders into each session.

Jaeden Schafer
Editor in Chief · · 4 min read
DigitalOcean launches Cuttlefish, an iPhone app for supervising AI agents

DigitalOcean launched Cuttlefish, a native iPhone app that lets developers watch and control AI agents running on its Managed Agents platform, announcing it at BoxWorks with Box as launch partner. The app streams live agent sessions, including replies, reasoning and tool calls, and lets users approve or deny tool calls, steer a run or stop it from a phone.

Cuttlefish is available on the App Store for Managed Agents customers, and DigitalOcean announced it on Nov 5, 2026. Managed Agents itself is in public preview. DigitalOcean said Cuttlefish works with whichever agent harness a team runs on the platform rather than being tied to one vendor's agent.

“Agentic work is a new paradigm. Builders direct agents that work on projects and check in when a decision needs a human”
— Vinay Kumar, Chief Product and Technology Officer, DigitalOcean

The supported harnesses are Claude Code, Codex CLI, OpenCode, OpenHands, Hermes, LangGraph and CrewAI. DigitalOcean's blog post says Managed Agents pairs those harnesses with the models and more than 20,000 tools suited to a given job, and that users can start a new session from the app by picking a coding agent.

Key facts

  • 01DigitalOcean launched Cuttlefish, a native iPhone app for its Managed Agents platform, on Nov 5, 2026.
  • 02Cuttlefish works with Claude Code, Codex CLI, OpenCode, OpenHands, Hermes, LangGraph and CrewAI harnesses.
  • 03Tool calls classified as high risk require Face ID or a passcode before approval, per DigitalOcean.
  • 04Box Mount, in private preview, syncs a Box folder into each Managed Agents session with Box permissions applied.
  • 05Managed Agents is in public preview, with billing that follows active CPU use.

Cuttlefish sends notifications when a run needs approval, finishes or fails. Approval screens show the full command, file path or diff the agent wants to act on, and DigitalOcean says requests classified as high risk require Face ID or the phone's passcode before they go through.

Beyond approvals, the app lets users send a correction, pause, resume or stop a session, and queue messages while the agent works. Users can also view websites and documents an agent has built, browse workspace files, attach to running servers, create checkpoints, and fork or roll back a session.

Sessions run in DigitalOcean MicroVMs that persist independently of any device, and each session gets its own isolated sandbox, according to the company. A teammate can join or take over a session when its owner is away. DigitalOcean said billing follows active CPU use, so time an agent spends waiting carries minimal cost; it did not publish rates.

Box's piece of the launch is Box Mount, which syncs a Box folder into each Managed Agents session with Box's permissions, version history and audit trail applied to every change. Box Mount is in private preview. In Cuttlefish, users connect it through the app's tools settings, sign in and pick a folder, and each person syncs with their own Box account. Edits and deletions made by the agent sync back to Box.

“With Box Mount, developers using DigitalOcean Managed Agents can give agents secure access to business content in Box”
— Biswaroop Palit, VP, Product Management Platform and Integrations, Box

Box describes Box Mount as a headless synchronization component that gives an agent sandbox a POSIX-compatible Box workspace, with Box remaining the governed system of record for permissions, versions, retention and audit history. Box said the tool is not meant to replace every Box API or MCP call.

Related · from this week
Anthropic launches message threads to cut Claude API agent latency
Jaeden Schafer · 4 min read →

DigitalOcean says it serves more than 680,000 customers. Its release notes that preview services are provided as is, are not intended for production use, and may change or be discontinued.

ShareXLinkedInEmail
AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

  • ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat
  • Generate images & video with Sora, Veo, Ideogram
  • Compare any two models side by side
  • From $8.99/mo · 80+ models, all included
Try AI Boxaibox.ai
Trusted by 3,000+ teams
Got a tip?

Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.

Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.

AI Box Daily briefingFree · Daily · No fluff

Stay ahead of everyone in AI.

The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.

Loved by 10,000+ AI professionals
Free forever. Unsubscribe with one click.

The briefing read inside teams at

Keep reading

More from Tools

Anthropic logo
Tools

Anthropic launches message threads to cut Claude API agent latency

The public beta lets agents send only new messages each turn; Anthropic says median time to first token falls about 35% on 500k–1M token requests.

Jaeden Schafer4 min read
Anthropic logo
Tools

Anthropic adds Claude Code rule against recreating specific existing games

The two-sentence prompt change on Opus 5.5 still allows original and genre games, but one licensed wargame developer says it is blocking months of work.

Jaeden Schafer4 min read
Anthropic logo
Tools

Karpathy's LLM wiki idea spawns a cottage industry of personal knowledge bases

A single April tweet from Andrej Karpathy sent thousands of researchers building Markdown wikis maintained by Claude and GPT-5.5.

Jaeden Schafer5 min read