Alstead Alstead · Answers
How do you see what an AI agent did?
To see what an AI agent did, give every agent its own key and record each operation in a log that is only ever added to. Each row should say who did it, what the operation was, and how it turned out. In Alstead, that log is the Ledger, and every operation has a name on it.
Why is it hard to tell what an agent did?
When several agents share one login or one API key, every action looks like it came from the same place. If a note changes or a record goes missing, you cannot tell which agent did it, or whether it was you. Chat history does not fix this either: it shows what an agent said, not every operation it ran.
The fix starts before anything goes wrong. Each agent needs an identity of its own, and every operation needs to be written down as it happens.
How does Alstead attribute each operation?
In Alstead, every agent has a name, a job and a key of its own. When you hire an agent, it gets its own access key. Alstead shows it to you once, then stores only a hash and a short prefix.
That prefix is what ties operations back to the agent. A Ledger row might read: Harbor, key prefix bzl_7QxK, memory.add, done. You can filter the Ledger by who: everyone, you, or a specific agent.
Because keys belong to one agent, you also control them one at a time. Pausing an agent blocks its key immediately. Rotating the key disables the old one. Retiring an agent permanently disables its key.
What does the Ledger record, and what does it leave out?
The Ledger records who did the operation, what it was, and how it turned out. Only new rows are ever added.
It records details about the operation, never the content of messages, memories or keys. Rows store permitted details only, such as the key prefix and a status code. So you get an accountable history without copying your private content into a log.
Alstead's home page describes the Ledger as built and tested, with entry IDs appearing in Needs you.
What happens when an agent tries something it isn't allowed to do?
Tools you don't tick are refused. When an agent reaches past its permissions, the call is refused and a request lands in Needs you with the exact action, the resource, and the permission it was missing. The request also carries an attribution ID, so you can match it to the Ledger.
You can allow that one action or deny it. Allowing adds exactly one permission to that agent's key, and Alstead checks the policy again when the agent retries. Deny is verified; allow is built but not yet live-verified. Payments, publishing and card use can't be approved at all in this release.
Can you see who changed your agent's memory?
Yes. You can search memory by text, section, tag, or which agent wrote it. When you forget something, it disappears from search, but the record of who wrote it stays. Your profile sections keep a version history, so you can see how "About you" changed and who changed it.
For more on scoping what each agent can reach, read agent permissions and always-on agent boxes, and for why smarter models still need limits, see smarter agents, same blast radius. For the bigger picture, see what is Alstead and what is a plane-first agent workspace.
FAQ
How can I tell which agent made a change? Give each agent its own key. In Alstead, every Ledger row carries the agent's name and key prefix, so you can filter by agent and see each operation it ran.
Does the Ledger store my messages or memories? No. It records details about each operation, never the content of messages, memories or keys.
Can entries in the Ledger be edited? Alstead writes each operation to a log that is only ever added to. Only new rows are ever added.
What do I see when an agent is blocked? A request in Needs you showing the exact action, the resource, and the missing permission. You can allow that one action or deny it.
Is Alstead available now? Alstead is an invite-only web alpha, free while in alpha.