Skip to main content
Buckle Up

​
Upgraded attachments & files

We’ve heard the complaints: too many unsupported file types, and poor extraction of attachment contents. To fix that, we built a document extraction service tuned for SIERRA. The new service will:
  • Support many image formats
  • Extract text from image/scan-style PDFs
  • Extract tables more accurately
  • Create captions/descriptions for embedded images and infographics
  • Convert all extracted content to Markdown format—ideal for AI comprehension
To deploy this upgrade, we’ll take these steps soon:
1

Finalize infrastructure deployment for the production Documents Service

2

Upgrade the Vercel AI SDK that SIERRA is built on

3

Replace the existing API calls for thread attachments

4

Replace the existing API calls for workspace files

5

Improve the SIERRA agent loop so it references attachments consistently

6

Migrate existing attachments and files to the new service

There will be backward compatibility issues. Expect a few new bugs — tell us what you hit and we’ll triage ASAP. We’ll likely also purge attachments that aren’t being used anymore; if something you need disappears, please let us know.

​
Documents Service API

Although built primarily for SIERRA, the Documents Service will be available via API for other ABS apps and solutions. For now the focus is on comprehensive, multiformat extraction into Markdown. If you use the service and want “style” adjustments to extracted content, we’re open to discussing improvements so it serves the organization, not just SIERRA.
In the future we expect the Documents Service and Tango to be further integrated, enabling both generic extraction and specialized workflows from the same endpoint. TBD.

​
Web search agents

After the SIERRA AI SDK upgrade, we’ll release a web search tool so SIERRA can look up non-proprietary, non-customer-derived content on the internet. If a query might be questionable, SIERRA will ask you to confirm before searching.

​
SIERRA API

We have a backlog of changes for the SIERRA API. The current API spec deviates from the OpenAI standard, which makes integrations painful for apps that expect OpenAI-compatible connectors. In the next API release we’ll remove those deviations. This will:
  • Make it easier to pass conversation threads to the API
  • Include GPT-5 models

​
API authentication

Today, SIERRA API keys are managed from the profile navigator (upper right) in SIERRA. To streamline things, key management will move to the AI Portal site so you can manage keys for all AI Center of Excellence endpoints from one place.

​
HR knowledge repository

We have big plans for specialized institutional knowledge repositories in SIERRA. SkillShare was our first step — giving SIERRA access to staff resumes and capabilities. Next up: HR manuals. The HR agent in SIERRA will answer questions using HR manuals specific to the user (accounting for location and company). This will be our second repo, but not our last. We aim to increase the cadence of repo releases rapidly. If you have ideas for proprietary tools or knowledge bases SIERRA should know about, please… let us know.

​
Last note

Twice recently, we have experienced major Microsoft Azure outages where SIERRA stopped working. It was them, not us. That said, the releases above may be bumpy. Thanks for joining us for the ride.