Issues

• Qwen3.5 templates drop earlier reasoning, so the next turn rereads the previous turn's tool results

  ashhart/TensorFold#3289 · Filter by author Whq02 opened on Nov 14, 2026

• qwen27 on Metal: with --thinking-reserve, a reasoning reply never counts as short, so srpt lets a long prompt take the GPU and the reply's decode drops ~95 %

  ashhart/TensorFold#3270 · Filter by author CodingEmil opened on Nov 13, 2026

• Unable to add stdio customer mcp server

  bug
  ashhart/TensorFold#3267 · Filter by author rjcdoc opened on Nov 13, 2026

• Responses API: accept namespace tools (Codex CLI sends its MCP and multi-agent tools as namespaces)

  ashhart/TensorFold#3266 · Filter by author tiziano-contorno opened on Nov 13, 2026

• Prompt cache eviction: measure on agent-style traces, then keep resumed sessions and reused prefixes

  ashhart/TensorFold#3241 · Filter by author ashhart opened on Nov 13, 2026
  4

• Grammar as a non-binding draft hint: tool-call envelopes drafted without masking

  ashhart/TensorFold#3237 · Filter by author ashhart opened on Nov 13, 2026
  2

• Docs: a Claude Code guide that keeps its prompt prefix cacheable

  ashhart/TensorFold#3225 · Filter by author ashhart opened on Nov 13, 2026
  1

• Images: progress events and previews while a picture generates

  ashhart/TensorFold#3224 · Filter by author ashhart opened on Nov 13, 2026
  2

• Images: n from 1 to 4 per request, each with its own seed

  ashhart/TensorFold#3223 · Filter by author ashhart opened on Nov 13, 2026
  2

• Grammar compile time: bound it, and keep one slow schema from blocking every grammar request

  ashhart/TensorFold#3209 · Filter by author ashhart opened on Nov 13, 2026
  4

• Qwen3.8-27B on Metal: context copies for a lone stream beside DFlash2

  ashhart/TensorFold#3201 · Filter by author ashhart opened on Nov 13, 2026
  4

• Lane host status reads that never stall a decode round

  ashhart/TensorFold#3200 · Filter by author ashhart opened on Nov 13, 2026
  1

• tensorfold launch <agent>: point a coding agent at the local server without touching its settings

  ashhart/TensorFold#3199 · Filter by author ashhart opened on Nov 13, 2026
  1

• Slow readers: bound a stream's undelivered events and pause its lane

  ashhart/TensorFold#3198 · Filter by author ashhart opened on Nov 13, 2026
  1

• Forget one conversation's saved states and keep the rest

  ashhart/TensorFold#3197 · Filter by author ashhart opened on Nov 13, 2026
  1

• Prompt cache salts: cache_salt and per-key isolation of kept states

  ashhart/TensorFold#3196 · Filter by author ashhart opened on Nov 13, 2026
  1

• Request log with replay and crash quarantine

  ashhart/TensorFold#3188 · Filter by author ashhart opened on Nov 13, 2026
  2

• Responses API: background requests (create, poll, cancel)

  ashhart/TensorFold#3187 · Filter by author ashhart opened on Nov 13, 2026
  2

• Logprobs with tools, stop strings, reasoning and the Responses API

  ashhart/TensorFold#3186 · Filter by author ashhart opened on Nov 13, 2026
  2

• structural_tag: constrain tagged regions inside free text

  ashhart/TensorFold#3184 · Filter by author ashhart opened on Nov 13, 2026
  4

• Admin route to drop the prompt cache's kept states

  ashhart/TensorFold#3183 · Filter by author ashhart opened on Nov 13, 2026
  7

• Post-filter probabilities: report the distribution a sampled token was drawn from

  ashhart/TensorFold#3182 · Filter by author ashhart opened on Nov 13, 2026
  9

• v0.6.6 TP2 on 2 x 16 GB: shard while loading and a tighter KV estimate (patch)

  ashhart/TensorFold#3179 · Filter by author Nipale-ai opened on Nov 13, 2026
  4

• docs: how tool calls are constrained under tool_choice and TENSORFOLD_TOOL_GRAMMAR

  ashhart/TensorFold#3177 · Filter by author ashhart opened on Nov 13, 2026
  2

• Flash Next on CUDA: warm turns, so the next turn resumes past the reply

  ashhart/TensorFold#3176 · Filter by author ashhart opened on Nov 13, 2026
  1