Changelog
What we shipped,week by week.
New features, improvements and deprecations across Inference, Agents and Evals.
RSS · JSON feed · status.lattis.dev
v2.4
Latest
Streaming tool calls
Tool calls now stream partial arguments and results, cutting perceived latency on multi-step runs by about 40%.
$ lattis run support-triage
streaming
plan
0.1 s
search_tickets
0.3 s
read_logs
0.4 s
issue_refund
awaiting approval
1,284 tokens · 1.2 s · trace tr_8f2a
All releases
Read release:
Streaming tool calls
Tool calls now stream partial arguments and results, cutting perceived latency on multi-step runs by about 40%.
Read release:
CI gate for GitHub Actions
A first-party action fails the build when a change scores below your threshold. Two lines of YAML.
- uses: lattis/evals-action@v1
with:
threshold: 0.90
Read release:
Semantic cache, second generation
Hit rate up from 31% to 54% on conversational workloads. Enabled per key.
Read release:
Regions: Frankfurt and Singapore
Data residency per API key. Latency from Berlin drops to 19 ms p50.
eu-central · Frankfurt
ap-southeast · Singapore
+ 9 more regions
Read release:
Agents 2.0: persistent sandboxes
Filesystem and processes persist for 24 hours between steps. Replays for every run.
Read release:
Legacy completions endpoint
/v1/complete is deprecated and will be removed Jan 31 2027. Use /v1/chat.
- POST /v1/complete
+ POST /v1/chat
The Log