Open low-latency framework for building voice and vision AI agents.
Vision Agents by Stream is an open framework for building low-latency voice and vision AI agents over WebRTC, using Stream's edge network. It works with any model or video provider, keeping the stack flexible. Apache-2.0 licensed, it is built for real-time conversational agents where latency matters.
| Pricing | Open Source |
|---|---|
| Category | Automation & Agents |
| Platforms | web, api |
| Company | Stream |
| Official website | github.com |
Automation and agent tools promise to do work for you, so judge them on reliability before cleverness. Map the exact workflow you want to automate, then check the tool's integrations cover every app in that chain — one missing connector breaks the whole flow. Look for clear triggers, sensible limits, and logs you can read when something goes wrong, because something eventually will. For agents that act on your behalf, approval controls matter: you should be able to review messages, purchases or changes before they happen, especially at the start. Pricing usually scales with runs or tasks, so estimate your monthly volume honestly. Start with one low-risk workflow, measure the time it truly saves, and expand from there.
Vision Agents by Stream is an open framework for building low-latency voice and vision AI agents over WebRTC, using Stream's edge network. It works with any model or video provider, keeping the stack flexible. Apache-2.0 licensed, it is built for real-time conversational agents where latency matters.
Vision Agents offers Low-latency voice and vision agents over WebRTC, Edge network for real-time performance, Model and video-provider agnostic and Apache-2.0 licensed.
Vision Agents is open source, so the software itself is free to download, use and self-host. Your only costs are the infrastructure you run it on and any paid services you connect to it.
Strengths of Vision Agents include Purpose-built for real-time agent latency; Backed by Stream's communications infrastructure. On the downside, Tied to Stream's edge network for best performance; Narrower scope than general agent frameworks.
by Ghosthand
Lets AI agents control a real iPhone from a Mac — screenshots over USB, taps/swipes/typing over Bluetooth — with no jailbreak, Xcode or app on the phone.
Best for: Developers and agent builders who want an AI agent to operate a real iPhone from a Mac
Pricing: Freemium
Limitation: Requires a Mac and a real iPhone — Linux, Windows and the iOS Simulator are not supported
by Pomerado
Turns any website with no API into a deterministic API/MCP for your agent — it builds and tests the integration itself, handles logins and 2FA, and self-repairs when the site changes.
Best for: Teams connecting agents to legacy portals and internal tools that have no API
Pricing: Freemium
Limitation: Free and $50 tiers allow only 1 account per website — unlimited accounts need the Contact-us plan
by boat
Persistent cloud Ubuntu VMs built for agent fleets — SSH, Docker, a 60fps virtual desktop and fast forking from snapshots, billed per second with stopped machines costing nothing.
Best for: Teams running fleets of coding agents that need real dev machines without managing infrastructure
Pricing: Paid
Limitation: $20/month plan minimum — the plan equals $20 of VM time, with overage paid from your wallet
by Agent Relay
Lets your team's agents talk to each other directly — a frontend agent can ask the backend agent what an endpoint returns, read each other's sessions and hand off work — instead of routing every handoff through a human copy-pasting between chats.
Best for: Teams running multiple agents that need to hand off work to each other directly
Pricing: Open Source
Limitation: The public homepage gives few technical details — setup runs through an agent-guided onboarding flow
by Relote
One shared memory for a person's AI tools and team: Claude, ChatGPT, Cursor and Claude Code save decisions, people and files to it over MCP and read them back next session
Best for: People and teams whose AI tools need to share decisions, clients and project memory
Pricing: Freemium
Limitation: Hosted service — your memory lives on Context Engine's servers, with an EU option at setup
by Monish
Free iPhone app that puts AI agents on the lock screen: live progress in the Dynamic Island, a ping when work lands, and Approve/Deny taps before an agent deploys, sends or spends
Best for: Developers who run agents overnight and want approvals on their lock screen
Pricing: Free
Limitation: iPhone only — the phone app is iOS, with no Android version stated
