Help users move faster through familiar workflows while retaining the review, approval, and operational processes those workflows require.
AI at Astra
Human judgment.
Amplified.
AI helps people work with information and complete tasks. We invest in people using AI to extend their expertise and capacity. Our engineers own the code they produce, and users decide which application changes to approve.
Human judgmentAt the center.Always in the loop.
01 / HUMAN + AI
You set the direction. Intelligence starts with the outcome you want to achieve.
Our conviction
Better tools. The same responsibility.
Our stance
A culture decision.
Then a technology decision.
How we think about AI shapes how we engineer, what we deliver, and how people interact with our products.
AI is a force multiplier for people. Human expertise remains essential to delivery, and our investment is in helping the team learn and use the tools well.
For customers, that means helping capable teams accomplish more with their available time and resources. Faster workflows leave more room for work that requires judgment, while speed, reliability, human focus, and trust guide the capabilities we build.
Useful AI connects to the application, its data, and the task at hand. Ground assistance in relevant information, validate proposed changes, and make outputs straightforward to review.
AI enhances human judgment. Engineers remain accountable for the code, and operators retain authority over decisions. Users set the intent, refine the proposed work, and decide when to act.
Security is central to AI-assisted work. Validation, identity, and explicit approval shape how actions move from a suggestion to a change inside an application, with existing permissions respected.
Our vision
From asking questions
to moving work forward.
Our direction is to extend governed AI actions across the applications people already use, with users directing the work and approving changes. The workflow below describes that application-level approach.
From requests to action
Bring AI-powered tasks
to existing applications.
Fabrial connects AI to the tools and workflows of a specific application. Your people keep using that application, its data, and its permissions. Integration happens through an application gem and the existing API, without moving the workflow into a separate data platform.
Your appChat + existing API
FabrialTools · Drafts · Approval
Your AIProvider of choice
Approved changes return through your application’s API.
- Lightweight deployment
- A single containerized service connects the application and its AI provider.
- Provider flexibility
- Choose commercial AI providers or compatible self-hosted models to suit the environment.
- Disconnected environments
- Use a compatible self-hosted provider within an on-premises or disconnected deployment.
- 01
Describe the task
Describe what you want to accomplish in the application you already use. Your task and application data provide the context.
- 02
Prepare a draft
AI prepares proposed work and checks it against application rules and reference data. Issues can be corrected before anything is saved.
- 03
Review & approve
Inspect the proposed change, ask for corrections, and approve it when ready. Changes wait for your explicit approval before they are saved.
- 04
Apply with your permissions
Approved changes go through the application using your identity and permissions. The application’s own access controls still apply.
- 05
Verify the result
Confirm that the saved result matches the approved work and record the action for review.
How Fabrial governs the work
Fabrial is a containerized orchestration service between an application and its AI provider. It supplies tool contracts, manages drafts, enforces approval, and calls the application’s API under the user’s delegated identity. The model receives tool contracts and results, not user credentials or direct application access.
Fabrial equips AI with gem-defined tools, JSON-Schema contracts, and live reference catalogs. Work takes shape in a server-side draft workspace outside the model’s context; validation errors return to AI for correction. Mutating operations wait at a confirmation gate for explicit approval.
Approved operations use delegated, role-scoped user identity through the application’s API. The application’s own authorization still applies. Fabrial reads back the saved result and records the operation for audit, connecting the request to its outcome.
Application gems
Define the gem.
Connect the application.
Tools, guided prompts, knowledge topics, reference catalogs, and commit rules live in application-specific gems. New integrations add a gem for the target application and its API.
Provider flexibility
Choose the model.
Fit the environment.
Fabrial supports adapters for OpenAI, Anthropic, Palantir Foundry, and self-hosted models through compatible serving infrastructure, including disconnected deployments.
Governed execution
Bounded access.
Accountable actions.
Schema and live-catalog validation, explicit confirmation, delegated user permissions, integration-scoped state, and audit records govern the path from a draft to a saved result.
AI in our engineering
More room for
the work that matters.
We use AI to reduce repetitive work and give engineers more space for architecture, security, performance, and sound design decisions.
From textbooks and manuals to search, Stack Overflow, and agentic AI, the tools available to engineers continue to evolve. Knowledge of the codebase, the domain, and the mission remains essential.
Equip the engineer.
We invest in people learning the tools. Pair domain understanding with new ways to develop and explore solutions, strengthening the team’s expertise and delivery capacity.
Own the output.
Engineers remain accountable for every AI-assisted outcome. Review, understand, and validate the work with the care expected of every engineering contribution.
Keep the mission first.
AI-assisted work follows the same security, review, and engineering standards as every other contribution. Evaluate tools against the environment’s operational constraints, with warfighter trust and mission outcomes at the center.
The next challenge starts with a conversation