Reliable infrastructure for the agentic era.
WeekndLabs builds open-source infrastructure for AI agents: a gateway for the credentials they hold, a context layer for what they read twice, and routing for the models they call. All of it under MIT and Apache 2.0, with 390 stars on GitHub. Every number we publish is measured on a real corpus and replays on yours.
One stack under your agent
An agent needs credentials for the services it calls, a way to stop paying for what it has already read, and a model to run on. Each layer is its own product and runs without the other two.
Context
Omni
v0.7.3Your agent pays twice for output it has already seen. OMNI hands back a retrievable handle instead. Nothing deleted, nothing invented, and every number replays on your own history.
Sits in front of Files and command output.
brew install fajarhide/tap/omniomni.weekndlabs.com- off a file the agent reads twice
- 97.2%
- off file reads across the corpus
- 89.6%
Credentials
Selat
v0.1.2Every agent you write ends up holding somebody else's OAuth. Selat holds it instead: connect GitHub, Google or Discord once from a browser, and your agent keeps one bearer that never rotates under it. MCP and REST see the same catalog, so Claude Desktop and a runtime you wrote yourself call the same tools.
Sits in front of GitHub, Google, Discord.
npx @fajarhide/selatselat.weekndlabs.comModels
Heimsense
v0.1.3Point Claude Code at any LLM. A Go proxy that routes your agent traffic to whichever model you actually want to pay for.
Sits in front of Any OpenAI-compatible model.
390 stars, 13 contributors and 90 releases across Selat, Omni, Heimsense and AI PR Describer, read from the GitHub API every hour. A count we cannot read in full is left out.
How it pays for itself
Run it yourself
The build you download is the whole product, on your own machine, with no account and no paid edition behind it. Read the commitment.
Or we run it
Selat also runs hosted for teams who would rather not operate a gateway, priced on tool calls rather than seats. GitHub Sponsors is the other channel.
Also from the lab
Dev workflow
The work around the agent, from the pull request it opens to the assumption you have not tested yet.
Desktop and foundations
The tokens the rest is built on, and the one that runs on your Mac.
Applications
What the infrastructure is for: a product people use, built on the Claude API.
Work with us.
Partnerships, integrations and investment conversations start with an email to [email protected].