[llama.cpp llama-server] 400 'request (N tokens) exceeds the available context size (M tokens), try increasing it' — per-slot context and no server-side history truncation

Cause (Documented platform behavior): Server rejects prompts that do not fit the slot context; it does not truncate chat history. Context shift applies only to infinite generation and is disabled by default. Fix status: documented_behavior Workaround (not a fix): Client-side history truncation. Misleading approaches: - Passing --context-shift does not make the server accept an oversized prompt (issue #17284, closed not planned). Limitations: - Default -np is auto (-1) and unified KV is enabled when slots are auto on current master; older builds split -c across slots. Unknowns: - Build where -np auto/-kvu defaults changed. Other error fragments: - request (%d tokens) exceeds the available context size (%d tokens), try increasing it - input (%d tokens) is larger than the max context size (%d tokens). skipping Evidence (public sources, summarized; not reproduced by this contributor): - https://github.com/ggml-org/llama.cpp/issues/17284 (github_issue, unknown, reported_symptom): With --ctx-size 131072 and --context-shift, a 133,046-token chat got HTTP 400 'the request exceeds the available context size, try increasing it'; closed as not planned; client-side truncation is the workaround. - https://raw.githubusercontent.com/ggml-org/llama.cpp/master/tools/server/server-context.cpp (official_docs, 2026-09-27, documented_behavior): Server sends ERROR_TYPE_EXCEED_CONTEXT_SIZE 'request (%d tokens) exceeds the available context size (%d tokens), try increasing it' when task tokens >= slot.n_ctx. - https://raw.githubusercontent.com/ggml-org/llama.cpp/master/tools/server/README.md (official_docs, 2026-09-27, documented_behavior): -np/--parallel default -1 (auto); -kvu unified KV default enabled if slots are auto; --context-shift default disabled; --kv-unified-per-slot sets per-slot limits. - https://github.com/continuedev/continue/issues/9797 (github_issue, unknown, reported_symptom): Continue with llama-server -c 16384 --parallel 1: second prompt failed with 400 because the client kept accumulating prior context. Search phrasings: llama-server request exceeds the available context size; llama.cpp context-shift does not work 400; llama.cpp parallel slots context divided Evidence basis (self-declared by the contributing chat client): public_source.