dev-agent

dev-agent commented on pull request dev/nisse#49

2026-09-27 19:05:23 +00:00

Benchmark seven new RTX 4090 models, and local-fast on the 4090

Two follow-ups measured after the runs:

Context probe (after the unittest key's rate limit was raised to 400k/min): every model's own server refused a prompt at 1.5× its declared window,…

dev-agent created pull request dev/nisse#49

2026-09-27 18:44:15 +00:00

Benchmark seven new RTX 4090 models, and local-fast on the 4090

dev-agent created branch bench-4090 in dev/nisse

2026-09-27 18:43:56 +00:00

dev-agent pushed to bench-4090 at dev/nisse

2026-09-27 18:43:56 +00:00

82f171c2f5 Benchmark seven new RTX 4090 models, and local-fast on the 4090

dev-agent created branch example-opus-5-5 in dev/nisse

2026-09-25 20:22:51 +00:00

dev-agent pushed to example-opus-5-5 at dev/nisse

2026-09-25 20:22:51 +00:00

3cb70c0656 Point the example config at Opus 5.5, with Haiku as consultant

dev-agent created pull request dev/nisse#48

2026-09-25 20:22:50 +00:00

Point the example config at Opus 5.5, with Haiku as consultant

dev-agent pushed to benchmark at dev/nisse

2026-09-25 01:25:16 +00:00

0cccfe1e94 Add the local baseline: qwen3-8b-awq and qwen3-14b, 5 runs per task

dev-agent commented on pull request dev/nisse#47

2026-09-25 01:25:14 +00:00

Add an executor benchmark that keeps its results

Baseline added: both local models, 5 runs per task, default tuning.

dev-agent created pull request dev/nisse#47

2026-09-25 00:35:25 +00:00

Add an executor benchmark that keeps its results

dev-agent created branch benchmark in dev/nisse

2026-09-25 00:35:25 +00:00

dev-agent pushed to benchmark at dev/nisse

2026-09-25 00:35:25 +00:00

50b46821d7 Add an executor benchmark that keeps its results

dev-agent opened issue dev/nisse#46

2026-09-25 00:21:12 +00:00

Separate architect and planner roles

dev-agent created branch enforce-tool-calls in dev/nisse

2026-09-24 22:54:52 +00:00

dev-agent pushed to enforce-tool-calls at dev/nisse

2026-09-24 22:54:52 +00:00

b934c46714 Enforce the tool-call contract, asking servers only where it works

dev-agent created pull request dev/nisse#45

2026-09-24 22:54:50 +00:00

Enforce the tool-call contract, asking servers only where it works

dev-agent commented on pull request dev/nisse#44

2026-09-24 13:38:17 +00:00

Fit every model request to its context window

It's used for one thing: the output ceiling sent with each request is the smaller of llm.max_output_tokens (our config, default 16384) and the model's max_output_tokens from LiteLLM. The…

dev-agent created branch context-budget in dev/nisse

2026-09-24 00:26:32 +00:00

dev-agent pushed to context-budget at dev/nisse

2026-09-24 00:26:32 +00:00

494715efa9 Fit every model request to its context window

dev-agent created pull request dev/nisse#44

2026-09-24 00:26:31 +00:00

Fit every model request to its context window