Skip to content

5 · Your first conversation

Open the workspace, pick the engine you sealed in step 3, and ask it something.

A reply means the whole chain worked: you signed in as the owner, the agent authorised you, it asked the router for a model, the router authorised that too, and the answer came back encrypted.

Four moving parts, and each fails with its own message. Check them in this order — the later ones cannot work if an earlier one is wrong.

Terminal window
hexaeight-activate restart agent
[licence] personal licence - N policy rule(s) in force
[identity-agent] <name> listening on :8770

0 policy rule(s) means owner is missing from hexaeight-agent.json, or the baseline policy was never written. Go back to step 2.

2. Is the router up, and did the agent find it?

Section titled “2. Is the router up, and did the agent find it?”

The router must be running before the agent. If it was not, restart the agent now.

A licence error on the agent when the licence is fine almost always means one of these two things:

  • the router was not running when the agent started
  • the command was run from the wrong folder, so env-file was never read
Terminal window
hexaeight-activate list-policy --subject [email protected]

This prints what the enforcer actually decides, not what is written down. A rule that exists but does not match reads as protection and is not.

The two rows people miss:

rulewithout it
[email protected],op:*,outboundyou sign in, and nothing else works
*,<agent-name>,outboundengines cannot reach the router at all

The router authorises twice — may this caller relay through this agent, and may this caller use this model. Both live in the router’s own policy, which the agent cannot write and which is read once at startup.

If you changed it, restart the router.

Read the last twenty lines, or the whole file. Do not search it for the word you expect — the line that explains a failure usually sits beside the one you searched for, and filtering hides it.

Terminal window
tail -20 my-agent/agent.log

An agent with a cryptographic identity, reachable only by you, that can reason using a model whose credentials it does not hold.

Nothing else is configured. It has no documents, no procedures, and no callers other than you — each of those is something you add deliberately.


you want toread
let it answer from your own documentsDocuments
put an API you already run behind itAn existing API
make it follow the same procedure every timeA procedure it follows
let another system call itOther software calling it
understand what you just builtArchitecture