Agentic AI > Agent Assist > Environment and commands reference

Environment and commands reference¶

This page provides a quick reference for environment variables, how the assist LLM is chosen, important files and directories, the list models prompt command, and slash commands.

If you use Agent Assist through a third-party coding agent instead of dr assist, install the skill with:

npx ai-agent-skills install datarobot-oss/datarobot-agent-skills

See the Agent Assist skill page for installation, setup, and usage guidelines.

Environment variables¶

The variables that can be set in the environment or in .env.

Variable	Required	Description	Default / notes
`DATAROBOT_API_TOKEN`	Yes (unless provided by DR CLI configuration)	DataRobot API key for LLM gateway (default provider).	—
`DATAROBOT_ENDPOINT`	No	DataRobot API endpoint.	`https://app.datarobot.com/api/v2`
`AGENT_ASSIST_LLM_BASE_URL`	No	Base URL for the assist session Completions API. See Change the Agent Assist LLM.	Derived from `DATAROBOT_ENDPOINT` (for example, `…/genai/llmgw`)
`AGENT_ASSIST_LLM_MODEL_NAME`	No	Model for the assist session. See Change the Agent Assist LLM.	`anthropic/claude-sonnet-4-5-20250929`
`AGENT_ASSIST_LLM_API_KEY`	No	API key for an external Completions API provider.	Falls back to `DATAROBOT_API_TOKEN` if unset—even for an external provider. See the warning below.
`AGENT_ASSIST_DISABLE_LLM_GATEWAY`	No	Skip the LLM gateway catalog fetch and route the assist session directly to the model named by `AGENT_ASSIST_LLM_MODEL_NAME` (for example, a DataRobot deployment). Requires `AGENT_ASSIST_LLM_BASE_URL` and a non-default `AGENT_ASSIST_LLM_MODEL_NAME` to also be set.	Unset (LLM gateway catalog used)
`LOGFIRE_TOKEN`	No	Logfire token for tracing.	—
`DATAROBOT_CLI_CONFIG`	No	Override path to DR CLI configuration file.	Default: `~/.config/datarobot/drconfig.yaml`
`AGENT_ASSIST_CONFIG`	No	Override path to Agent Assist configuration file.	Default: `~/.config/datarobot/agent_assist_config.yaml`

Files and directories¶

The paths and files used or created by DataRobot Agent Assist. All paths are relative to the current working directory when you start dr assist.

Run dr assist in an empty directory

Only run dr assist from a dedicated and empty directory. Running this command in a directory containing code or other files is unsafe. When you use the agent assist coding workflow, the assistant clones the DataRobot Agent Application Template repository into the current directory. This action can overwrite or conflict with existing files, damaging the existing project and degrading the accuracy of the assistant's output. Before running dr assist, if you're not in a dedicated directory, create one and open the terminal there (for example, mkdir my-agent && cd my-agent, then run dr assist).

Path	Description
`agent_spec.md`	Agent specification file (YAML content) in the current working directory; written by the assistant during design. See Agent specification reference for field definitions and examples.
`.env`	Optional environment file in the current directory; same variable names as above.
`.datarobot/cli/versions.yaml`	Used by dependency check; defines minimum tool versions.
`~/.config/datarobot/drconfig.yaml`	DR CLI configuration (token, endpoint). This path can be overridden with `DATAROBOT_CLI_CONFIG`.
`~/.config/datarobot/agent_assist_config.yaml`	Agent Assist configuration (optional LLM base URL, model name, API key). This path can be overridden with `AGENT_ASSIST_CONFIG`.
`~/.config/datarobot/agent_assist/settings.yaml`	Global DataRobot CLI / Agent Assist preferences (for example language). Does not set the assist LLM model; use `AGENT_ASSIST_LLM_MODEL_NAME` in the environment or `.env`. For more information, see Change the Agent Assist LLM.
`config.yaml`	Optional repository-level configuration; can set `repository` (`url`, `branch`, `target_dir`, `tag`) for the template clone.

`list models` command¶

At the $ prompt, type list models (two words, no leading slash) to print the Available LLM Gateway Models list from your DataRobot LLM gateway. Agent Assist shows a table with a row index, Model Name (for example azure/gpt-5-1-2025-11-13 or bedrock/anthropic.claude-sonnet-4-5-20250929-v1:0), Provider (such as Anthropic, Azure OpenAI, Amazon Bedrock, Google Vertex AI, or Together AI), and a short Description. After the table, the output includes how many models are listed, a reminder of which model the session is configured to use, and a prompt to either change the model or continue the conversation.

The exact models and providers depend on your organization’s gateway configuration and entitlements. Which model the session uses is determined as described in Change the Agent Assist LLM.

Apart from list models and the slash commands below, use natural language at the $ prompt for design, coding, and deployment tasks.

Slash commands¶

Built-in session commands that start with / at the $ prompt. Typing / alone lists these commands and suggests /help for details. Simulation session commands apply only inside simulation, not at the main $ prompt.

Command	Alias	Description
Core commands
`/help`	`/?`	List commands or show help for a command: `/help [command]`.
`/quit`	`/exit`	Exit the session (with confirmation).
Session commands
`/reset`	`/new`	Clear session state and start fresh (removes `agent_spec.md` and template directory, clears conversation).
`/checkpoint`	-	List or restore session checkpoints.
`/compact`	-	Compact conversation history to reduce token usage.
`/resume`	-	Show the resume context injected into the current session.
`/save`	-	Save session state to disk.
Simulation session commands
`/quit`	`/exit`	Exit the simulation and return to the main session (no confirmation prompt).
`/f`	-	Record feedback about the design: `/f <text>`. Compiled into a summary when the simulation ends, which the assistant uses to help refine `agent_spec.md`.

Change the Agent Assist LLM¶

Agent Assist uses two LLMs:

Assist session: For design, simulation, and coding conversations. Set AGENT_ASSIST_LLM_* variables to use the DataRobot LLM gateway or an external Completions API.
Coded agent: For the application Agent Assist generates. Set the model field in agent_spec.md to an LLM gateway model ID on your deployment. External AGENT_ASSIST_LLM_* settings do not apply to the coded agent.

Assist session¶

Set AGENT_ASSIST_LLM_MODEL_NAME in the environment or in a .env file in the directory where you run dr assist. When more than one value is set, Agent Assist uses the first match in this order:

Environment variable in the shell.
.env in the current working directory.
Built-in default: anthropic/claude-sonnet-4-5-20250929.

LLM gateway (default). Leave AGENT_ASSIST_LLM_BASE_URL unset. Set AGENT_ASSIST_LLM_MODEL_NAME to a model ID from list models (for example anthropic/claude-sonnet-4-5-20250929).

If the default model isn't in your gateway catalog

The built-in default (anthropic/claude-sonnet-4-5-20250929) may not be enabled for your organization's LLM gateway. If dr assist can't reach that model, check which models your organization has enabled from a terminal—no running session required—using the DataRobot CLI:

dr llm list

Copy a value from the MODEL column (for example azure/gpt-5-1-2025-11-13), then set it before you start dr assist:

export AGENT_ASSIST_LLM_MODEL_NAME=azure/gpt-5-1-2025-11-13

dr llm list is a shorthand alias for dr llm-gateway list. It lists the same active gateway catalog as the in-session list models command, plus any DataRobot-deployed LLMs available to your account. See the llm-gateway command reference for the full flag and output reference.

If a model isn't in the catalog at all, an org admin must enable it; see LLM availability.

Use an external LLM for Agent Assist

Set AGENT_ASSIST_LLM_BASE_URL, AGENT_ASSIST_LLM_MODEL_NAME, and AGENT_ASSIST_LLM_API_KEY for any provider that exposes an OpenAI-compatible chat completions endpoint. Set all three—if you omit AGENT_ASSIST_LLM_API_KEY, Agent Assist falls back to sending your DATAROBOT_API_TOKEN to this external endpoint instead of failing. Add the variables to .env or export them before you run dr assist:

Example: Anthropic

AGENT_ASSIST_LLM_MODEL_NAME=claude-sonnet-4-5
AGENT_ASSIST_LLM_BASE_URL=https://api.anthropic.com/v1
AGENT_ASSIST_LLM_API_KEY=YOUR_ANTHROPIC_API_KEY

Use model names and URLs from the provider documentation. The list models command lists gateway models only.

Coded agent¶

The model field in agent_spec.md must be an LLM gateway model ID. Gateway models must be available in your organization's LLM gateway before you run the Code an AI agent workflow. See the Agent specification reference.

Model settings

Consider the following when changing agent assist LLM settings:

~/.config/datarobot/agent_assist/settings.yaml stores CLI preferences (for example language). It does not set AGENT_ASSIST_LLM_MODEL_NAME.
Restart dr assist after you change AGENT_ASSIST_LLM_* variables. There is no in-session /model command.
Credential precedence is described in Set configuration variables.