diff --git a/dasha/.claude/settings.json b/dasha/.claude/settings.json
new file mode 100644
index 0000000..e53cd2a
--- /dev/null
+++ b/dasha/.claude/settings.json
@@ -0,0 +1,10 @@
+{
+  "$schema": "https://json.schemastore.org/claude-code-settings.json",
+  "env": {
+    "ANTHROPIC_BASE_URL": "http://localhost:4000",
+    "ANTHROPIC_MODEL": "qwen3-8b",
+    "ANTHROPIC_DEFAULT_SONNET_MODEL": "qwen3-8b",
+    "ANTHROPIC_DEFAULT_OPUS_MODEL": "gemma3-27b",
+    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "qwen3-4b"
+  }
+}
diff --git a/dasha/.claude/settings.local.json.example b/dasha/.claude/settings.local.json.example
new file mode 100644
index 0000000..3c61a95
--- /dev/null
+++ b/dasha/.claude/settings.local.json.example
@@ -0,0 +1,5 @@
+{
+  "env": {
+    "ANTHROPIC_AUTH_TOKEN": "dummy"
+  }
+}
diff --git a/dasha/.gitignore b/dasha/.gitignore
new file mode 100644
index 0000000..37505ac
--- /dev/null
+++ b/dasha/.gitignore
@@ -0,0 +1,2 @@
+.claude/settings.local.json
+.env
diff --git a/dasha/README.md b/dasha/README.md
new file mode 100644
index 0000000..49c0139
--- /dev/null
+++ b/dasha/README.md
@@ -0,0 +1,25 @@
+# Dasha Compute · red de Macs Apple-silicon (alpha)
+
+[Dasha Compute](https://www.getdasha.com/compute) es una red de Macs reales que sirven modelos locales detrás de una API OpenAI-compatible. No expone endpoint Anthropic nativo: se intercala LiteLLM como proxy traductor (ver `_docs/proxies.md`).
+
+- **Base URL (OpenAI-compat)**: `https://www.getdasha.com/compute/api/v1`
+- **Guest key (24h, sin registro)**: `curl -X POST https://www.getdasha.com/compute/api/guest-keys` → el campo `api_key` de la respuesta (`dgk_...`)
+- **Modelos**: la red sirve un conjunto pequeño y cambiante (al escribir esto: `qwen3-4b`, `qwen3-8b`, `gemma3-12b`, `gemma3-27b`); la lista viva está en `GET /compute/api/v1/models`
+- **Estado**: alpha. La disponibilidad varía con los Macs conectados; no hay SLA.
+- **Docs**: <https://www.getdasha.com/compute/llms.txt>
+
+## Uso
+
+```bash
+pip install 'litellm[proxy]'
+export DASHA_GUEST_KEY=dgk_...   # la key que devuelve /compute/api/guest-keys
+litellm --config litellm-config.yaml --port 4000
+cp .claude/settings.local.json.example .claude/settings.local.json
+claude
+```
+
+El `model_name` de cada entrada del config de LiteLLM coincide con los valores `ANTHROPIC_MODEL` / `ANTHROPIC_DEFAULT_*` de `.claude/settings.json` (mapeo por tamaño: haiku → qwen3-4b, sonnet → qwen3-8b, opus → gemma3-27b). Cambia los nombres si la lista viva de modelos cambia.
+
+## Costo
+
+La guest key es gratuita (24h, chat + lista de modelos). Es una red comunitaria en alpha: sirve para probar el flujo, no para producción.
diff --git a/dasha/litellm-config.yaml b/dasha/litellm-config.yaml
new file mode 100644
index 0000000..58b31d7
--- /dev/null
+++ b/dasha/litellm-config.yaml
@@ -0,0 +1,27 @@
+# Dasha Compute via LiteLLM (proxy traductor - ver _docs/proxies.md)
+# 1) Mint a 24h guest key:  curl -X POST https://www.getdasha.com/compute/api/guest-keys
+#    (response field: api_key, prefix dgk_)
+# 2) export DASHA_GUEST_KEY=dgk_...
+# 3) litellm --config litellm-config.yaml --port 4000
+# Live model list: https://www.getdasha.com/compute/api/v1/models
+model_list:
+  - model_name: qwen3-4b
+    litellm_params:
+      model: openai/qwen3-4b
+      api_base: https://www.getdasha.com/compute/api/v1
+      api_key: os.environ/DASHA_GUEST_KEY
+  - model_name: qwen3-8b
+    litellm_params:
+      model: openai/qwen3-8b
+      api_base: https://www.getdasha.com/compute/api/v1
+      api_key: os.environ/DASHA_GUEST_KEY
+  - model_name: gemma3-12b
+    litellm_params:
+      model: openai/gemma3-12b
+      api_base: https://www.getdasha.com/compute/api/v1
+      api_key: os.environ/DASHA_GUEST_KEY
+  - model_name: gemma3-27b
+    litellm_params:
+      model: openai/gemma3-27b
+      api_base: https://www.getdasha.com/compute/api/v1
+      api_key: os.environ/DASHA_GUEST_KEY