Get Started
What we maintain
The models, providers, integrations, tools and security behind the agent are Alloy's job. Your engineers work on the robot, not on the agent.
When you build your own debugging agent, your team owns everything around it: the models, the data connections, the tools, the evals and every fix when something breaks. When you use Alloy, that is our job.
Your engineers work on the robot. We keep the agent working.
What Alloy maintains for you
| Area | What we do | What your team does |
|---|---|---|
| Models | Track new models, test each one on robotics tasks, and move each job to the model that tests best. See Model testing. | Nothing. You choose how deep the agent goes, not which model it uses. |
| Model providers | Run the agent across more than one provider, with failover, so one provider's outage does not stop an investigation. Keep data-residency settings in place through every change. | Nothing. |
| Integrations | Build and maintain the connections to Slack, Jira, Confluence, GitHub, Grafana, ClickHouse, S3 and Google Cloud Storage. Fix them when a vendor changes its API. | Connect once. |
| Custom sources | Forward-deployed engineers connect your in-house APIs, tools and log formats with you, and keep them working. | Give us a scoped credential. |
| Log formats | Support MCAP, ROS bags, PX4 and ArduPilot logs and telemetry, and add new formats with you. See Log formats. | Send a sample. |
| The agent's tools | Maintain the analysis sandbox, the robotics tools and the instructions the agent runs on, and test every change against our evals before it ships. | Nothing. |
| Rollouts | Stage every model and tool change, organisation by organisation, because no test suite catches everything. | Nothing. |
| Reliability | Monitor the platform around the clock and run a defined incident process. Ask for the incident-response summary in the security pack. | Tell us if something looks wrong. |
| Security | Keep SOC 2 Type II, organisation isolation and access controls current. See Security and trust. | Review the security pack once. |
What it costs a team to own this itself
Teams that build their own debugging agent on a general model usually start fast. The cost arrives later:
- More than one agent. Each team builds its own version. Soon nobody knows which one is reliable, what data each one can see, or who to call when one breaks.
- Model churn. A new model ships every few weeks. Someone has to test it on your robot data before switching, or the answers quietly get worse.
- Broken connections. Ticketing, chat and storage APIs change. A connector that worked last quarter fails without notice.
- No memory. A personal setup remembers for one engineer. The rest of the team starts from zero. See The knowledge base.
- Your best engineer becomes the maintainer. The person who knows the robot best spends their time keeping the tool alive.
With Alloy, the software is the delivery mechanism. What you buy is the outcome: investigations that work, kept working by a team whose job it is.
Someone accountable
When something goes wrong, you talk to the people who built it. Your Alloy engineers know your setup, your data and your workflows, and they own the fix. You have named engineers who own the fix.
Related
The knowledge base
Every investigation adds to your team's shared knowledge base, so a known issue gets a known answer and what one engineer works out helps everyone.
Rollout and time to value
Value in week one, often on day one. How Alloy's forward-deployed engineers switch the agent on with you and help the whole team adopt it.