Database tools

Database tools let the agent run real queries against your data — not a copy, not a snapshot — so it can answer questions about today's orders, this week's signups, or the customer's current balance.

Shared safety model

Every database tool ships with three layers of protection:

1. Your DB role — the database user the connection string points to. Provision it with least privilege (a read-only replica, a role with SELECT only).
2. Capability profile — read-only by default on every new credential. Read+write enables INSERT / UPDATE; full access enables DELETE. Enforced before the query reaches the DB.
3. Hard caps — 500 rows per response, 256 KB serialized, 30 s timeout, DDL (DROP/CREATE/ALTER) rejected unconditionally, multi-statement SQL rejected.

PostgreSQL

Tool id: postgres_query. Credential type: postgres.

Use list_schemas, list_tables, or describe_table as discovery shortcuts, then run parameterised SQL with $1, $2 placeholders.

Setup

1. Create a least-privilege role in Postgres: CREATE ROLE chatzuri_agent LOGIN PASSWORD '…'; GRANT CONNECT ON DATABASE app TO chatzuri_agent; GRANT USAGE ON SCHEMA public TO chatzuri_agent; GRANT SELECT ON ALL TABLES IN SCHEMA public TO chatzuri_agent;
2. Save the connection details — either a full URI (postgresql://user:pass@host:5432/db?sslmode=require) or the discrete host / port / database / username / password / sslMode fields.
3. Pick the SSL mode — use require for cloud-hosted Postgres (Supabase, Neon, RDS).

MySQL / MariaDB

Tool id: mysql_query. Credential type: mysql.

Same shape as Postgres but with ? placeholders. Works for self-hosted MySQL, MariaDB, PlanetScale, and AWS RDS.

Setup

1. Create a least-privilege user: CREATE USER 'chatzuri'@'%' IDENTIFIED BY '…'; GRANT SELECT ON app.* TO 'chatzuri'@'%'; FLUSH PRIVILEGES;
2. Save connection details on a MySQL credential. Set SSL mode to require on managed providers (PlanetScale, RDS).

MongoDB

Tool id: mongodb_query. Credential type: mongodb.

Actions: find, findOne, aggregate, countDocuments, distinct, list_collections, describe_collection, plus inserts / updates / deletes gated by the profile.

Setup

1. Get the connection string from MongoDB Atlas (Database → Connect → Drivers) or your self-hosted cluster.
2. Make sure the user in the URI has only the roles you want — e.g. read on a specific database for a read-only agent.
3. Save the URI + optional defaultDatabase on a MongoDB credential.
4. On Atlas, allow chatzuri's outbound IPs (or use 0.0.0.0/0 for a public test cluster). Production: lock to your VPC peer.

Redis

Tool id: redis_query. Credential type: redis.

24 commands across read (GET, MGET, EXISTS, TYPE, TTL, HGET, HGETALL, LRANGE, SCAN, …), write (SET, HSET, LPUSH, ZADD, EXPIRE, INCR, …), and delete (DEL, HDEL, SREM). FLUSHDB / FLUSHALL / KEYS * are never reachable regardless of profile.

Setup

1. Get connection details from your provider (Upstash, AWS ElastiCache, self-hosted). Use a rediss:// URI for TLS, or set the TLS field on discrete-field credentials.
2. Set a namespacePrefix (e.g. chatzuri:agent:abc:) — the tool prefixes every key the LLM supplies and limits SCAN to {prefix}*. Strong multi-tenant isolation; recommended for shared Redis.

Supabase

Tool id: supabase_query. Credential type: supabase.

Talks to your Supabase project's PostgREST endpoint — select, insert, update, upsert, delete, rpc, list_tables, describe_table.

Setup

1. Open your project in app.supabase.com → Project Settings → API.
2. Copy the Project URL.
3. Pick a key: anon key (subject to Row-Level Security — recommended) or service_role key (bypasses RLS — full access).
4. Save URL + key + optional defaultSchema (defaults to public) as a Supabase credential. Set the capability profile.

Airtable

Tool id: airtable. Credential type: airtable.

Actions: list_tables, list_records, get_record, create_record, update_record, delete_record.

Setup

1. Open airtable.com/create/tokens.
2. Click Create new token. Add scopes: data.records:read at minimum, plus data.records:write and schema.bases:read if you want writes and list_tables.
3. Add access to the specific base(s) you want the agent to touch.
4. Click Create token and copy the personal access token (starts with pat…).
5. Save the token + base id (visible in the base URL, https://airtable.com/{base_id}/…) as an Airtable credential.