Agent Runtime

RealTimeX now ships a fuller runtime layer around agents. Instead of only choosing an agent and sending a prompt, you can decide which tools it can use, which folders it can touch, when it should run, and how you review what happened afterward.

Where to find it

• Settings > Agentic CLIs
• Settings > Agents > Working Directories
• Settings > Agents > Ambient Agent
• Agent feed in the main sidebar
• Settings > Diagnostics on desktop

Agentic CLIs

Agentic CLIs are external command-line tools your agents can rely on, such as gh or glab.

• RealTimeX tracks whether each CLI is installed, authenticated, and ready.
• Admins can add CLIs from the registry or register custom executables.
• RealTimeX does not store those CLI credentials for you. You authenticate the CLI in your terminal, then RealTimeX probes readiness and exposes that status to agents.
• If the launched runtime itself needs a provider API key, use Agent Authentication instead of assuming CLI login state is enough.
• This is separate from Credentials, which stores reusable secrets for agent workflows and integrations.
• You can attach install docs, auth commands, probe commands, and a short prompt hint so the agent knows when the CLI should be used.

Agentic CLIs are not the same thing as CLI agents. CLI agents are runtimes such as Antigravity CLI, Cursor Agent, Codex CLI, or Claude Code. Agentic CLIs are the tools those runtimes use to do work.

Typical flow

1. Install and authenticate the CLI on the machine running RealTimeX.
2. Add it in Settings > Agentic CLIs.
3. Probe it to confirm install and auth state.
4. Enable it so agents can rely on it during sessions.

For the full setup and troubleshooting guide, see Agentic CLIs.

Terminal CLI agents

RealTimeX also supports interactive terminal agent runtimes such as Antigravity CLI, Cursor Agent, Claude Code, Codex CLI, Gemini CLI, Qwen CLI, and OpenCode.

One important implementation detail is that not every runtime uses the same launch path:

• some runtimes can back ACP-style chat sessions
• some runtimes are terminal-first and should be launched from the terminal runtime instead

Antigravity CLI notes

Antigravity CLI is currently a terminal-first runtime in RealTimeX.

• RealTimeX recognizes the handles agy, antigravity, and antigravity-cli
• launch it from the terminal runtime path rather than ACP chat session creation
• authenticate it with agy or the Antigravity IDE
• RealTimeX can inspect the current Antigravity model for runtime status, but it does not switch Antigravity models at launch

If Antigravity appears installed but cannot be launched from an ACP session, that is expected behavior for the current product build.

Cursor Agent notes

Cursor Agent is ACP-launchable in RealTimeX and also exposes terminal runtime metadata.

• RealTimeX discovers it with the synthetic provider id acpx-cursor
• terminal launch resolves the agent command first and can fall back to cursor-agent
• set CURSOR_CLI_PATH if you need to force a specific Cursor CLI path
• Cursor uses its own authentication instead of an injected Agent Authentication API-key profile
• ACPX-backed Cursor sessions currently use approve-all; interactive approval prompts are not forwarded through RealTimeX yet

Agent Authentication

Agent Authentication stores provider API keys that RealTimeX injects into launched agent runtimes such as Claude Code, Codex CLI, Gemini CLI, Qwen CLI, OpenCode, and Ambient Agent.

• Use it when the runtime needs a provider key at launch time.
• This is separate from Agentic CLIs, which track tool readiness.
• This is also separate from Credentials, which store outbound integration secrets for workflows and tools.

Agents that authenticate through their own CLI or OAuth state instead of injected API keys, such as Antigravity CLI and Cursor Agent, do not primarily use this page.

For the full guide, see Agent Authentication.

Working Directories

Working Directories let you pre-authorize folders outside a workspace so agents can read and write there without repeated permission prompts.

• Add any trusted absolute path.
• Attach a short description so the agent can choose the right folder on its own.
• Optionally limit a directory to specific workspaces or specific agents.
• Add a wiki project from Personal Notes directly into the registry if you want agents to treat that project as a writable working area.

This is the safest way to let an agent work across real local projects without giving it unrestricted filesystem access.

Ambient Agent

Ambient Agent is the scheduled agent runner that wakes up in the background and acts on HEARTBEAT.md.

• Choose which agent runs and optionally override its LLM.
• Set a check interval, timezone, and active hours.
• Add scheduled task blocks inside HEARTBEAT.md so different jobs can run on their own cadence.
• Use custom instructions for tone, language, or priorities.
• Trigger a manual wake immediately, or queue instructions for the next scheduled run.
• Review recent activity, last result, and next wake-up from the same screen.

Calendar-aware routines

Ambient Agent can also add calendar context to its work:

• Morning brief
• Pre-meeting prep
• Follow-up suggestions

These routines are most useful when your HEARTBEAT.md instructions already explain what the agent should do with that context.

For the full scheduler, HEARTBEAT.md, and calendar routine guide, see Ambient Agent.

Agent feed

The Agent feed page is a live operations board for agent work.

• Watch tasks move through Upcoming, In Progress, Review, Done, Failed, and Canceled.
• Filter the board down to one or more agents.
• Open tasks that need user input or approval.
• Use it as the fastest way to see what your background and scheduled automation is doing right now.

Diagnostics

Desktop builds include a runtime diagnostics surface for monitored sessions.

• Inspect recorded sessions from the current machine.
• Review incidents, timeline events, and sampled app health.
• Open the raw session folder locally.
• Generate a diagnostic bundle or review a ticket draft when you need to escalate an issue.

Diagnostics is operational tooling, not a user chat log. Use it when you need to understand what happened around a runtime problem.

Best practices

• Keep Agentic CLI entries focused on tools agents actually need.
• Register only the directories you truly trust.
• Treat HEARTBEAT.md as the source of truth for what your Ambient Agent is allowed to do.
• Use Agent feed for active work and Diagnostics for post-mortem analysis.