diff options
| author | Dmitry Vyukov <dvyukov@google.com> | 2026-01-20 15:03:18 +0100 |
|---|---|---|
| committer | Dmitry Vyukov <dvyukov@google.com> | 2026-01-21 13:38:45 +0000 |
| commit | c9820ab0fe2dce0914ff01bcaf3829ca82150eb2 (patch) | |
| tree | 350f597291bec42753754dc1948acf8166c5ba68 /pkg/aflow/execute.go | |
| parent | f1d5c3ecdec0b86db1df926ccd3553157988690d (diff) | |
pkg/aflow: cache LLM requests
Using cached replies is faster, cheaper, and more reliable.
Espcially handy during development when the same workflows
are retried lots of times with some changes.
Diffstat (limited to 'pkg/aflow/execute.go')
| -rw-r--r-- | pkg/aflow/execute.go | 3 |
1 files changed, 3 insertions, 0 deletions
diff --git a/pkg/aflow/execute.go b/pkg/aflow/execute.go index 19f8d3aec..3e1a6a112 100644 --- a/pkg/aflow/execute.go +++ b/pkg/aflow/execute.go @@ -178,6 +178,9 @@ func (ctx *Context) generateContentGemini(model string, cfg *genai.GenerateConte return nil, fmt.Errorf("model %q does not exist (models: %v)", model, models) } if thinking { + // Don't alter the original object (that may affect request caching). + cfgCopy := *cfg + cfg = &cfgCopy cfg.ThinkingConfig = &genai.ThinkingConfig{ // We capture them in the trajectory for analysis. IncludeThoughts: true, |