Skip to content
Almostly Almostly Almostly

About Almostly

Almostly builds building blocks, not frameworks: small, open-source tools for data, ML, and AI engineering you can read in a sitting and bend to your own work. For engineers, researchers, and hackers who want to see how things work, not adopt a black box. Local-first, and grown out of years of building high-impact machine learning in fintech and banking.

What we believe

Section titled “What we believe”
  • Primitives, not frameworks. Composable pieces you wire together yourself, not an opinionated framework you have to adopt wholesale.
  • You stay in control. The decisions that matter (the model, the data, the loop) stay yours. The tool removes the busywork around them, nothing more.
  • Readable by design. Code you can read in one sitting. If you can’t understand it, you can’t trust it or hack it.
  • Local-first & open source. Runs on your machine, against local or hosted models. Everything’s public: fork it, read it, change it.

What we build

Section titled “What we build”
wrencode A minimal agent harness for coding, in a single Python file. The whole tool-calling agent loop fits in one readable file, runs against local or hosted models, interactively or headless, and is yours to hack.
oblako A local AWS platform. Run Bedrock, SageMaker, Redshift, Step Functions and more on your laptop, with real engines behind the AWS APIs and your boto3 code unmodified.
pgsesame Permissions as code for PostgreSQL and Amazon Redshift. Declare roles, users, groups and grants in YAML, then plan and apply the difference like Terraform.

More small, sharp tools are on the way. Browse everything on GitHub.

Who’s behind it

Section titled “Who’s behind it”

Almostly is built by its founding members.

Get in touch

Section titled “Get in touch”