Use this file to discover all available pages before exploring further.
These examples show the intended feel of hosted agents: a user describes an outcome, the agent gathers context, uses the right tools, and turns the result into a reviewable next step. Some examples combine API resources that are available now with dashboard experiences that are still being productized.
Deploy github.com/acme/helpdesk to staging. It needs Postgres and should be reachable at helpdesk-staging.example.com.
2
The agent inspects the repository
It reads the project files, detects the runtime, finds required environment variables, and identifies the database dependency.
3
The agent asks for missing inputs
It renders a short form for the domain, environment variables, and deployment target instead of asking you to write a manifest.
4
The agent creates a reviewable change
It prepares the package, deployment configuration, and install changes, then asks for approval before applying anything sensitive.
Conversation
You: Deploy this repository to staging.Agent: I found a Node app with a Postgres dependency. I can create a staging install,generate the deployment config, and set up the database connection.[Form: deployment target, domain, missing environment variables]Agent: I prepared a repository change request with the package config and install values.[Approval: Create staging install]
Agents can use external data tools when the decision depends on live market data. Akua’s Hetzner Value Auctions tool indexes dedicated server auctions, CPU benchmarks, memory, disks, locations, and price so an agent can recommend hardware before a user buys it.
Conversation
You: I need a cheap worker for Postgres. At least 64 GB RAM, NVMe preferred, under 50 EUR/month.Agent: I’ll compare current auction servers against that workload.Agent: Best fit right now:1. 64 GB RAM, NVMe storage, strong single-thread score, under your budget. Good fit for Postgres and small multi-tenant workloads.2. 128 GB RAM option, slightly more expensive. Better if you expect large indexes or want more cache headroom.I recommend option 1 for the current workload. Want me to add it as a workercandidate for your cluster, or save it for later review?[Choice: Add worker candidate] [Save recommendation] [Show details]
The user gets a recommendation in the same deployment flow. The agent can explain the tradeoff in plain language, or show CPU models, disk layout, datacenter, hourly cost, and benchmark scores for advanced users.
You: Why did the customer install fail after the last update?Agent: I’ll compare the current install, recent change request history, deployment events,and logs.Agent: The install started failing after the latest values change removedDATABASE_URL from the runtime environment. There is already a stale change request thatadds it back and updates the secret reference.[Card: Existing change request found][Choice: Continue change request] [Create new change request] [Only summarize]
This is why agents need read access to existing repository change requests. They can avoid duplicate work, continue a previous attempt, or explain why a change request should be replaced.
Ambient agents are useful when the right time to ask for help is before a human opens the chat.
Ambient event
Signal: production install failed twice in 10 minutesAgent: I started a read-only investigation for the failed production install.Findings:- The new version is healthy in staging.- Production fails only in one region.- The failing pods cannot pull the image from the private registry.Suggested next step:Rotate the registry token for the affected workspace and retry the rollout.[Approval: Create repository change request] [Open install] [Dismiss]
The agent is always available, but it does not need to consume expensive runtime resources while waiting. It starts work only when configured signals match policy.
