Bundled Terminal and CLI

#One Package

The daita command is bundled with daita-agents. There is no separate CLI or client package to install.

bash

pip install daita-agents
daita

With no subcommand, daita opens the interactive terminal application. Resume a known agent directly with:

bash

daita --agent atlas

Use --root before the subcommand to select an alternate state root:

bash

daita --root /absolute/path/daita-state sources atlas

#Agent and Source Management

bash

daita create atlas
 
daita attach atlas sqlite ./warehouse.db --source-name Warehouse
daita attach atlas files ./exports --source-name Exports
daita attach atlas postgresql \
  --host db.example.com \
  --database analytics \
  --username readonly_agent \
  --password-env POSTGRES_PASSWORD \
  --schema public \
  --ssl-mode require \
  --source-name Analytics
 
daita sources atlas
daita detach atlas source-id --yes

PostgreSQL schemas can be repeated with multiple --schema arguments. The default port is 5432, the default schema is public, and the default SSL mode is require.

Detaching a source removes the registration and Daita-owned credentials; it never changes source data.

#Run and Chat

Run one request:

bash

daita run atlas "Summarize revenue by region" \
  --model openai:gpt-5.6-terra

Continue a conversation or emit lifecycle events:

bash

daita run atlas "Now compare it with last quarter" \
  --model openai:gpt-5.6-terra \
  --conversation-id conversation-id \
  --events-jsonl

Start an interactive chat:

bash

daita chat atlas --model openai:gpt-5.6-terra
daita chat atlas --model openai:gpt-5.6-terra --conversation conversation-id

Model IDs use provider:model. A custom compatible endpoint can be supplied to run with --base-url. Unreviewed model identities must first be admitted through the interactive terminal with explicit context-window and output limits.

#Interactive Commands

Inside chat, /help lists the current commands. Common entries include:

text

/status
/conversation
/new
/resume <conversation-id>
/sources
/learn <material>
/memory
/review
/user
/skills
/<skill-name> [request]
/exit

Ctrl-C interrupts the current request; Ctrl-D exits the interactive session.

#Memory, Learning, and Skills

bash

daita memory read atlas --target memory
daita memory edit atlas --target user
daita memory inspect atlas
daita memory review atlas --model openai:gpt-5.6-terra --cost-limit 0.05
daita memory list-candidates atlas --status awaiting_review
 
daita skills list atlas
daita skills show atlas monthly-revenue-review
daita skills save atlas monthly-revenue-review \
  --description "Compare monthly revenue." \
  --instructions-file ./monthly-review.md

See Memory, Reviewed Learning, and Skills for the underlying API contracts and approval behavior.

#Destructive Commands

The non-interactive commands require explicit confirmation:

bash

daita conversations clear atlas --yes
daita delete atlas --yes

Clearing conversations removes transcripts, candidate records, and review stamps but preserves approved knowledge. Deleting an inactive agent removes its agent home and Daita-owned credentials, never attached source data.