With a coding agent
Two plugins for Claude Code and Codex: one teaches the agent TessariDB, the other gives it a memory stored in TessariDB.
A coding agent writes TessariQL the way it writes any language it half knows: from memory, and confidently. The plugins in tessaridb/tessaridb-agent-plugins give it the language as it actually is, and they install the same way in Claude Code and in Codex.
| Plugin | What it gives the agent | Needs |
|---|---|---|
tessaridb | a skill covering TessariQL, every engine, transactions and history, the five clients, and running a node or a cluster | nothing — no server, nothing to start |
tessaridb-agent-memory | a memory that outlives the context window, stored in TessariDB, and a skill saying when to use it | the memory container running on the same machine |
Installing
In Claude Code:
/plugin marketplace add tessaridb/tessaridb-agent-plugins
/plugin install tessaridb@tessaridb
/plugin install tessaridb-agent-memory@tessaridb
In Codex:
codex plugin marketplace add tessaridb/tessaridb-agent-plugins
codex plugin add tessaridb@tessaridb
codex plugin add tessaridb-agent-memory@tessaridb
Install only the first if all you want is an agent that writes TessariQL correctly.
What the tessaridb skill is
Plain Markdown: a short SKILL.md that says when it applies and which file to
read, and a references/ folder with the detail — the language, search, the
other engines, transactions and history, clients, and operations. It names the
release it was checked against, and every TessariQL example in it has been run
against that release, so an example that stopped being
true is caught before an agent copies it.
It leads with the mistakes that succeed: a pooled connection that forgot its
USE, a value formatted into a statement instead of bound, a whole-record
UPDATE that drops the fields it did not mention, a script taken for a
transaction. Each of those runs without an error, which is why an agent needs to
be told.
When the skill and this site disagree, this site is right, and the skill says so.
Before the memory plugin works
The memory plugin registers a server; it does not carry one. It connects to
http://127.0.0.1:39142/mcp, where the memory container serves, so that
container has to be running first, and the agent is authorized once in the
browser. The repository's
README
has the docker run line and the sign-in steps for both hosts.
Until the container is up, the plugin is installed and the server shows as failing to connect. That reads like a plugin fault and is not one.
Codex signs in to the memory service as its own client, which the service knows
from 0.3.0-alpha on. With an earlier memory service, use the memory plugin
from Claude Code.