- Long Snowflake queries: an agent’s
run_sql_queryon Snowflake waits up to 60 seconds, then returns a query ID instead of timing out. The agent checks on it withget_sql_query_resultor stops it withcancel_sql_query, and queries still running are cancelled when the session stops. See Tools reference. - A tool call that runs past its 5-minute limit now says so, instead of retrying with an unclear error.
- Each agent’s slug shows as a chip you can copy in the agent list, the builder, and the session header.
- The Defer to picker in agent environment settings is searchable. A target that no longer exists shows as (not found).
- A new agent environment connection defaults its dbt™ target to
dev. - The “no warehouse connection” message now points to Settings > Environments > Agent, not to your own connection.
- Agent Slack tools that find or attach to a thread work only in the agent’s own channel.
- When saving an agent is blocked by a network filter, the builder says which text in the spec likely caused it, instead of “Failed to fetch”.
- Agent environments no longer show a “No connected MCP servers” message.
- The run summary’s DinoAI button reads View DinoAI fix (or View DinoAI fixes, with a count) when the run already has a fix session, instead of starting a new one. See Fix with DinoAI.
- dbt™ Fusion run logs show pass, warn, error, and skip counts with previous and next navigation. Fusion’s dim header lines are readable in dark mode.
- Deploy as YAML on a schedule made from a template now writes a real slug, and the preview shows it. Before, it could write the template’s timestamped label as the slug.
- A schedule whose
schedule_triggerorrun_afternames a schedule that doesn’t exist now shows an error after deploy. Before, it silently never ran. See Run completion trigger. - A
paradime dinoaicommand now waits up to 24 hours for its agent session, like other Bolt commands. Before, it failed after 1 hour while the agent kept working. - Fix with DinoAI in Slack works for companies that share one Slack workspace.
- The Bolt API has
boltRunNodeExecutions(runId): the result of every dbt™ node in a run, with status, timing, failures, and the adapter response, paged up to 500 at a time.
- dbt-base images are updated across every supported version: dbt™ 2.0.9, 1.12.6 (with dbt-clickhouse back), 1.11.16, 1.10.19, 1.9.8, and 1.8.8. The Code IDE, Bolt, and agents use the same images.