In a world where humans and agents are working together, its important for both to work within governed environments with guardrails. With this motivation, I am excited to talk about Environments in Paradime.

At Paradime, we are building the platform to automate data engineering using AI agents. We are giving our customers the picks and shovels to build, run, and observer data pipelines and agents from a single control plane.

To address growing challenges of managing and administering data platform across groups, teams, and domains, we invented Paradime Workspaces. Workspaces are isolated enclaves within your Paradime instance with its own users, repo, warehouse connections, and integrations.

But it had a problem.

Workspaces mapped to domains or departments with users ring-fenced between workspaces is the right model and it works.

However, each workspace could only connect to one code repository and the close coupling so that each user could have only one dbt™ version - was not fit for purpose anymore.

These constraints posed a significant limitation for both humans and agents when working within their workspace.

What are environments?

Environments are isolated infrastructure enclaves that can have their own warehouse, harness, dbt™ version and code repo(s). As a result each environment can have its own unique capabilities, and guardrails.

Users can create environments within each of the Paradime products so Code IDE, Bolt or Agents. As shown in the figure above, for the agent you can create 3 different environments as follows:

• Data Analyst - an environment that gives access to only certain Snowflake roles and warehouses so that when an agent is bound to that environment, the agent cannot work outside their access guardrails.

• Incident Manager - an environment that gives agents the power to solve data incidents as a result needs access to multiple repos and a different Snowflake role.

• Analytics Engineer - an environment that grants read on sources, write on views and marts.

Similarly for Bolt, the orchestration engine for DinoAI agents and data pipelines, users can have one environment for CI, another for testing new dbt versions, or an environment with just lightweight Python. The combinations are endless and ultimately built around giving the data engineer the toolkit to expand their imagination and build what’s possible in an agent-first world.

Conclusion

Environments is an exciting new abstraction on top of instances and workspaces giving you flexibility to customize the enclave and guardrails around how DinoAI agents and data pipelines operate.

That flexibility is a must have to orchestrate data pipelines and agents in a governed and durable way.

Learn more on environments in our docs and you can sign up to build your first agent with their own environment.