Blog
Stay up to date with the latest from AGNT5.
We ran identical legal document analysis tasks against Claude, OpenAI, and Gemini - with and without prompt caching. Here are the exact token counts, latency numbers, and cost math at every scale.
A walk through the Studio run view, the data sources behind it, and why the timeline is the artifact every other feature points back to.
How AGNT5's control plane separates projects from environments, and why that split keeps deploys boring.
Why AGNT5 models evaluation as just another kind of run — same journal, same replay, same observability.
How the AGNT5 runtime keeps one tenant's runs from affecting another — across journals, Parquet partitions, and the coordinator's routing table.
Every journal entry for an LLM step carries tokens_in, tokens_out, and cost_usd. Here is how that rolls up into a bill you can reason about.
Why the AGNT5 SDKs share a Rust implementation under the hood, what lives on which side of the FFI boundary, and what it costs.
How AGNT5 pauses a running workflow for a human decision — and why the durability model makes this a one-line primitive.
Entities give you stateful objects with exactly one writer at a time. Here is how that contract holds under crashes, retries, and concurrent callers.
Gateway, Engine, and Coordinator ship as one binary — and split apart when you need them to. Here is how the --target flag works, and why.
Embedding DuckDB inside the AGNT5 runtime to serve listings, filters, and analytics over S3 Parquet — without running a separate warehouse.
Why the AGNT5 runtime pushes completed runs to S3 as Parquet, and how that keeps the hot tier small and the query tier fast.
A walk through what happens when an AGNT5 run resumes after a crash — the journal, memoization, and the contract the SDK upholds.
The tradeoff space for a durable execution log, and why we landed on an embedded LSM over Kafka, Postgres, or a custom file format.
Why we're launching a blog and what to expect from future updates.
Five questions to ask before you ship an AI workflow to production.