YashShelar007

YashShelar007/convene

Run Claude Code headlessly as a local inference layer: named experts, multi-turn sessions, measured concurrency. Local and first-party only — no proxy, no network endpoint.

1 stars 0 forks 1 watchers Python MIT License
agents anthropic claude claude-code headless inference llm python
View on GitHub
6 Open Issues Need Help Last updated: Aug 28, 2026

Open Issues Need Help

View All on GitHub
Find the real concurrency ceiling (n=20 is just the largest burst tried) about 2 hours ago
help wanted measurement
YashShelar007/convene
1

Run Claude Code headlessly as a local inference layer: named experts, multi-turn sessions, measured concurrency. Local and first-party only — no proxy, no network endpoint.

Python
#agents#anthropic#claude#claude-code#headless#inference#llm#python
Measure Haiku's prompt-cache threshold (documented 2048, unverified here) about 3 hours ago
help wanted good first issue measurement
YashShelar007/convene
1

Run Claude Code headlessly as a local inference layer: named experts, multi-turn sessions, measured concurrency. Local and first-party only — no proxy, no network endpoint.

Python
#agents#anthropic#claude#claude-code#headless#inference#llm#python
How long does a prompt-cache entry survive on this path? about 3 hours ago
help wanted measurement
YashShelar007/convene
1

Run Claude Code headlessly as a local inference layer: named experts, multi-turn sessions, measured concurrency. Local and first-party only — no proxy, no network endpoint.

Python
#agents#anthropic#claude#claude-code#headless#inference#llm#python
Reproduce FINDINGS.md on Linux or Windows (everything so far is macOS arm64) about 3 hours ago
help wanted good first issue measurement
YashShelar007/convene
1

Run Claude Code headlessly as a local inference layer: named experts, multi-turn sessions, measured concurrency. Local and first-party only — no proxy, no network endpoint.

Python
#agents#anthropic#claude#claude-code#headless#inference#llm#python
Has --bare become the default for -p in any build after 2.1.237? about 3 hours ago
help wanted good first issue measurement
YashShelar007/convene
1

Run Claude Code headlessly as a local inference layer: named experts, multi-turn sessions, measured concurrency. Local and first-party only — no proxy, no network endpoint.

Python
#agents#anthropic#claude#claude-code#headless#inference#llm#python
What does hitting the subscription rate limit actually look like? about 3 hours ago
help wanted measurement
YashShelar007/convene
1

Run Claude Code headlessly as a local inference layer: named experts, multi-turn sessions, measured concurrency. Local and first-party only — no proxy, no network endpoint.

Python
#agents#anthropic#claude#claude-code#headless#inference#llm#python