diff --git a/README.md b/README.md
new file mode 100644
index 0000000..ed858ca
--- /dev/null
+++ b/README.md
@@ -0,0 +1,45 @@
+# prain
+
+Serve a mixture-of-experts model with vLLM when its experts do not fit in VRAM. A TOML file places
+each expert of each layer in VRAM (pinned or in a cache slot), in RAM, or in a pack file on disk.
+Missing experts are copied to the GPU or computed on the CPU (`[miss]`). Eviction, heat, prefetch
+and miss policies are configurable, and `prain tune` picks them from a routing trace.
+Target: Gemma 4 26B-A4B (30 layers x 128 experts, 42.5 GiB of experts) on an 8 GB RX 6600 XT.
+
+## Build (GPU host, user space)
+
+    scripts/build.sh wrappers sdk fetch torch torchvision vllm prain
+
+- `sdk`: venv with AMD's ROCm SDK wheels for `PRAIN_GPU_ARCH` (default gfx1032)
+- `torch`, `torchvision`: from source with clang and `-march=native` (hours)
+- `vllm`: v0.32.0 from source with gfx1032 added and `patches/vllm-*.patch` applied
+- `prain`: editable install and the test suite
+
+Check the stack: `scripts/smoke.sh MODEL_DIR --attention-backend TRITON_ATTN`
+
+## Serve
+
+    . scripts/env.sh
+    prain serve examples/gemma4-26b-a4b-rx6600xt.toml MODEL_DIR -- --served-model-name gemma4 \
+        --language-model-only --max-model-len 4096 --max-num-batched-tokens 1024 \
+        --enable-prefix-caching --attention-backend TRITON_ATTN
+
+The first start writes the pack (~3 min). Chat with `prain tui`, or `prain web --config FILE` for
+several users with drafts.
+
+## Tune for a machine
+
+    python scripts/measure.py PACK_DIR/layer029.w13 --threads 16 > cost.json   # server stopped
+    prain sim TRACE CONFIG MODEL_DIR/config.json --cost cost.json
+    prain tune TRACE MODEL_DIR/config.json --cost cost.json --expert-vram-gib 1.67 --ram-gib 0 \
+        --pack-dir PACK_DIR --cpu-threads 16 > tuned.toml
+    python scripts/bench.py http://127.0.0.1:8000
+
+The simulator treats every disk-tier expert as a disk read; the page cache makes real runs faster.
+
+## vLLM patches
+
+- `vllm-0.32-rocm-no-cuda-families`: gfx1032 reports capability 10.3, and vLLM's CUDA "family"
+  checks took it for Blackwell (B200 attention tiles: 128 KiB of LDS on a 64 KiB GPU).
+- `vllm-0.32-gemma4-fit-8gb`: Gemma 4 shares `embed_tokens` as `lm_head` instead of allocating a
+  second 1.4 GiB copy while loading, and sizes rope caches by `max_model_len` (saves ~0.4 GiB).
diff --git a/examples/gemma4-26b-a4b-rx6600xt.toml b/examples/gemma4-26b-a4b-rx6600xt.toml
index 33f1c4e..ee099e1 100644
--- a/examples/gemma4-26b-a4b-rx6600xt.toml
+++ b/examples/gemma4-26b-a4b-rx6600xt.toml
@@ -1,6 +1,9 @@
 # prain placement: prithivMLmods/gemma-4-26B-A4B-it-Uncensored-MAX on an 8 GB RX 6600 XT
 # Model: 30 MoE layers x 128 experts, top-8 routing, 11.34 MiB per expert (BF16).
-# VRAM: 8 GiB - 4.45 GiB dense weights - 0.75 GiB KV cache - ~0.9 GiB runtime = ~1.7 GiB experts.
+# VRAM, measured: 6.22 GiB weights (4.46 dense + 5 slots x 30 layers) + 0.75 GiB KV cache: 7.9 of
+# 8 GiB in use. Serve with --attention-backend TRITON_ATTN (ROCM_ATTN needs more LDS than RDNA2 has).
+# RAM tier: pinning is very slow behind a strict IOMMU (Thunderbolt eGPU: minutes per GiB), so this
+# file keeps no RAM tier; the kernel page cache holds the hot part of the pack instead.
 #
 # Tiers:  vram = pinned in a VRAM slot for the whole run
 #         ram  = pinned host memory, copied into a cache slot when a token needs it
@@ -38,10 +41,11 @@ budget = 2            # experts per layer and forward
 min_prob = 0.05       # skip guesses below this router probability
 
 [miss]                # an expert that is not in VRAM
-mode = "copy"         # copy | cpu (always compute it on the CPU) | auto (CPU if it has
-cpu_max_tokens = 1    #   at most cpu_max_tokens tokens, else copy)
-cpu_threads = 4       # CPU threads of the vLLM worker (sets OMP_NUM_THREADS)
-admit = 1             # cpu/auto: copy this many CPU-run experts into VRAM afterwards
+mode = "auto"         # copy | cpu (always compute it on the CPU) | auto (CPU if it has
+cpu_max_tokens = 2    #   at most cpu_max_tokens tokens, else copy). Measured on the Thunderbolt
+                      #   eGPU host: a copy costs ~4.6 ms + disk read, the CPU ~2.5 ms per token
+cpu_threads = 16      # CPU threads of the vLLM worker (sets OMP_NUM_THREADS)
+admit = 0             # cpu/auto: copy this many CPU-run RAM-tier experts into VRAM afterwards
 
 [serve]               # parallel requests (`prain serve` passes these to vLLM)
 max_requests = 2      # requests vLLM runs at once (--max-num-seqs); warm-ups count too