Tools Reference
The following tools can be specified in the tools.list field of an agent YAML. Use allowlist mode to restrict an agent to only the tools it needs.
Filesystem
read_file
Read the contents of a file in the repo
write_file
Write or overwrite a file in the repo
search_files_and_directories
List and navigate the repo file structure
ripgrep_search
Fast text search across files
Web
execute_web_search
Search the web for current information
extract_url_content
Fetch and read the content of a URL
perplexity_search
AI-powered web search via Perplexity
SQL & Terminal
run_sql_query
Execute a SQL query against the connected warehouse
run_terminal_command
Run a shell command (e.g. dbt™ commands)
get_snowflake_query_performance_stats
Deep-dive into a specific Snowflake query by ID
get_bigquery_query_performance_stats
Deep-dive into a specific BigQuery job by its ID
Orchestration
invoke_agent
Spawn a child agent defined in agents_squad
notify_parent_session
Wake a parent agent waiting on a callback
run_subagent
Run a lightweight sub-task without a full agent pod
Integrations
post_slack_message
Post a message to a Slack channel
get_linear_issue
Read a single Linear issue by ID (requires Linear integration)
list_linear_issues
List Linear issues filtered by team, assignee, workflow state, label or priority (requires Linear integration)
Bolt
The background agent can now both inspect and trigger / retry Bolt schedules and runs. Triggered runs go through Paradime's normal scheduler — custom commands are validated server-side before dispatch, so an agent cannot run arbitrary shell.
list_bolt_schedules
List active Bolt schedules in the workspace (name, UUID, cron, owner, commands). Use first to discover valid schedule_name_str values.
get_bolt_run_logs
Fetch AI-generated summaries of failed commands from a Bolt run. Pass run_id to inspect a specific run, or schedule_name_str to get the latest finished run of a named schedule.
trigger_bolt_schedule_run
Trigger a Bolt schedule run by name. Optional commands to override the schedule's commands (validated server-side); optional branch to override the git branch. Returns the new run_id.
retry_bolt_run
Retry a specific failed Bolt run by run_id. Creates a new run that resumes from the failed command; the original run is unchanged.
retry_latest_bolt_schedule_run
Retry the latest failed run of a Bolt schedule by name, without needing to know the run ID. Returns the new run_id.
Custom commands are gated. When an agent passes commands to trigger_bolt_schedule_run, each entry is checked against Paradime's allow-list of safe commands (is_allowed_command). Disallowed commands return an [ERROR] and the run is not dispatched.
Typical pattern: list_bolt_schedules → trigger_bolt_schedule_run (or retry_latest_bolt_schedule_run) → get_bolt_run_logs with the returned run_id to inspect outcome.
Data Warehouse
Warehouse tools are automatically available based on your workspace's connected data warehouse. Supported warehouses:
Snowflake
BigQuery
Redshift
Databricks
Trino
SQL Server
ClickHouse
Microsoft Fabric
Related
YAML Configuration — How to use
tools.modeandtools.listAgent-to-Agent Delegation — Using
invoke_agentandnotify_parent_sessionExamples — See tutorials on how to get started deploying agents
Related
YAML Configuration — How to use
tools.modeandtools.listAgent-to-Agent Delegation — Using
invoke_agentandnotify_parent_sessionExamples — See tutorials on how to get started deploying agents
Last updated
Was this helpful?