Pull requests · charmbracelet/fantasy

feat: add synthetic provider with quota-aware rate limits

#434LarsArtmann opened

Oct 24, 2026

Contributor1 of 2 tasks

feat: honor and surface rate-limit retry deadlines

#433LarsArtmann opened

Oct 24, 2026

Contributor1 of 2 tasks

fix(examples): sync structured output module dependencies

#430Ehsan-hash opened

Oct 22, 2026

fix(google): keep tool calls whose input is empty or malformed

#428pcbeingused333 opened

Oct 21, 2026

fix(jsonrepair): preserve escaped backslashes in strings

#427Ehsan-hash opened

Oct 20, 2026

1 of 2 tasks

fix(openai): keep streamed tool calls apart when the provider omits their index

#426pcbeingused333 opened

Oct 20, 2026

feat(providers): add opt-in gzip request bodies

#425andreynering opened

Oct 19, 2026

Member

fix(examples): replace retired Claude Sonnet 4 in web-search example

#424pcbeingused333 opened

Oct 18, 2026

test: deduplicate the agent and openai test suites

#422LarsArtmann opened

Oct 17, 2026

Contributor1 of 2 tasks

test: pin object streaming, content codecs, and openrouter reasoning formats

#421LarsArtmann opened

Oct 17, 2026

Contributor1 of 2 tasks

chore: enable the dupl linter and remove the duplication it reports

#420LarsArtmann opened

Oct 16, 2026

Contributor7 of 8 tasks

6

fix: return NoObjectGeneratedError with usage on object parse failures

#417LarsArtmann opened

Oct 16, 2026

Contributor5 of 7 tasks

fix(google): report streamed and cached Gemini usage correctly

#416someotherdustin opened

Oct 10, 2026

1 of 2 tasks

Stop dropping the first chunk of every OpenRouter reasoning block

#413h8mankind opened

Oct 6, 2026

Contributor

fix(openai): preserve Responses replay metadata and error details

#412kevinle128 opened

Oct 5, 2026

9 tasks done

fix(schema): preserve numeric and boolean enum types

#410takiAA opened

Oct 4, 2026

1 task done

fix(openai): preserve reasoning replay and check stream completion

#407kevinle128 opened

Oct 1, 2026

7 tasks done

2

fix(kronk): do not rewrite a terminal finish reason into tool-calls

#402feiiiiii5 opened

Sep 29, 2026

Contributor

2

fix(kronk): do not count cached prompt tokens in InputTokens

#401feiiiiii5 opened

Sep 29, 2026

Contributor

3

docs: add Tsubasa openaicompat example

#394cenab opened

Sep 27, 2026

fix(google): keep terminal finish reason when tool calls are present

#392feiiiiii5 opened

Sep 26, 2026

Contributor

2

feat: surface early provider metadata on the stream before content

#389andreynering opened

Sep 25, 2026

Member

fix(openai): treat GPT-6 as a reasoning model

#386kimnamu opened

Sep 24, 2026

1 of 2 tasks

1

fix(retry): lowercase header keys and surface Gemini RetryInfo

#363someotherdustin opened

Sep 12, 2026

1 task done

1

fix(openai): normalize streamed stop responses with tool calls

#360akirarat opened

Sep 11, 2026

1 of 2 tasks

Navigation Menu

Repository navigation

All pull requests

feat: add synthetic provider with quota-aware rate limits

feat: honor and surface rate-limit retry deadlines

fix(examples): sync structured output module dependencies

fix(google): keep tool calls whose input is empty or malformed

fix(jsonrepair): preserve escaped backslashes in strings

fix(openai): keep streamed tool calls apart when the provider omits their index

feat(providers): add opt-in gzip request bodies

fix(examples): replace retired Claude Sonnet 4 in web-search example

test: deduplicate the agent and openai test suites

test: pin object streaming, content codecs, and openrouter reasoning formats

chore: enable the dupl linter and remove the duplication it reports

fix: return NoObjectGeneratedError with usage on object parse failures

fix(google): report streamed and cached Gemini usage correctly

Stop dropping the first chunk of every OpenRouter reasoning block

fix(openai): preserve Responses replay metadata and error details

fix(schema): preserve numeric and boolean enum types

fix(openai): preserve reasoning replay and check stream completion

fix(kronk): do not rewrite a terminal finish reason into tool-calls

fix(kronk): do not count cached prompt tokens in InputTokens

docs: add Tsubasa openaicompat example

fix(google): keep terminal finish reason when tool calls are present

feat: surface early provider metadata on the stream before content

fix(openai): treat GPT-6 as a reasoning model

fix(retry): lowercase header keys and surface Gemini RetryInfo

fix(openai): normalize streamed stop responses with tool calls