diff --git a/dimos/agents/skills/speak_skill.py b/dimos/agents/skills/speak_skill.py
index b46de157c4..027019a781 100644
--- a/dimos/agents/skills/speak_skill.py
+++ b/dimos/agents/skills/speak_skill.py
@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import os
 import threading
 import time
 
@@ -28,6 +29,21 @@
 logger = setup_logger()
 
 
+def openai_api_key_set() -> str | None:
+    """Blueprint requirement check: SpeakSkill uses OpenAI text-to-speech, which needs
+    OPENAI_API_KEY. Returns None if set, else a clear message (mirrors ollama_installed).
+    Lets a missing key fail fast at blueprint build instead of crashing later in start()."""
+    if os.environ.get("OPENAI_API_KEY"):
+        return None
+    return (
+        "OPENAI_API_KEY is not set. The agentic blueprint uses OpenAI for text-to-speech "
+        "(SpeakSkill), and by default for the LLM agent too.\n"
+        "\n"
+        "   Set it with: export OPENAI_API_KEY=<your-key>\n"
+        "   Get a key at https://platform.openai.com/api-keys"
+    )
+
+
 class SpeakSkill(Module):
     _tts_node: OpenAITTSNode | None = None
     _audio_output: SounddeviceAudioOutput | None = None
diff --git a/dimos/robot/unitree/go2/blueprints/agentic/_common_agentic.py b/dimos/robot/unitree/go2/blueprints/agentic/_common_agentic.py
index 93312225bc..c51ef77bd5 100644
--- a/dimos/robot/unitree/go2/blueprints/agentic/_common_agentic.py
+++ b/dimos/robot/unitree/go2/blueprints/agentic/_common_agentic.py
@@ -15,7 +15,7 @@
 
 from dimos.agents.skills.navigation import NavigationSkillContainer
 from dimos.agents.skills.person_follow import PersonFollowSkillContainer
-from dimos.agents.skills.speak_skill import SpeakSkill
+from dimos.agents.skills.speak_skill import SpeakSkill, openai_api_key_set
 from dimos.agents.web_human_input import WebInput
 from dimos.core.coordination.blueprints import autoconnect
 from dimos.robot.unitree.go2.connection import GO2Connection
@@ -27,6 +27,11 @@
     UnitreeSkillContainer.blueprint(),
     WebInput.blueprint(),
     SpeakSkill.blueprint(),
+).requirements(
+    # SpeakSkill (OpenAI TTS) needs OPENAI_API_KEY — fail fast at blueprint build with a
+    # clear message instead of an ExceptionGroup deep in start_all_modules. Mirrors the
+    # ollama_installed gate used by the -ollama variant.
+    openai_api_key_set,
 )
 
 __all__ = ["_common_agentic"]
diff --git a/docs/installation/osx.md b/docs/installation/osx.md
index 8f190c1c5e..33e418be93 100644
--- a/docs/installation/osx.md
+++ b/docs/installation/osx.md
@@ -4,10 +4,12 @@
 # install homebrew
 /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
 # install dependencies
-brew install gnu-sed gcc portaudio git-lfs libjpeg-turbo python pre-commit
+brew install gnu-sed gcc portaudio git-lfs libjpeg-turbo python
 
 # install uv
 curl -LsSf https://astral.sh/uv/install.sh | sh && export PATH="$HOME/.local/bin:$PATH"
+# NOTE: the `export` above only affects the current shell. Open a new terminal so `uv` is on
+# PATH in later sessions — the installer adds it to your shell profile (e.g. ~/.zprofile).
 ```
 
 ## Using DimOS as a library
diff --git a/docs/installation/ubuntu.md b/docs/installation/ubuntu.md
index 60c5b986fe..f9b8fc5f5f 100644
--- a/docs/installation/ubuntu.md
+++ b/docs/installation/ubuntu.md
@@ -2,10 +2,20 @@
 
 ```sh skip
 sudo apt-get update
-sudo apt-get install -y curl g++ portaudio19-dev git-lfs libturbojpeg python3-dev pre-commit
+# Required system libraries. libturbojpeg + portaudio19-dev cover image/audio; libgl1 +
+# libegl1 are the OpenGL runtime for open3d and rerun-sdk (both always-installed core deps).
+# Without libgl1/libegl1 the visualizer fails at runtime with
+# "libGL.so.1: cannot open shared object file" (e.g. on minimal/headless/Docker installs).
+sudo apt-get install -y curl g++ portaudio19-dev git-lfs libturbojpeg libgl1 libegl1 python3-dev
+
+# optional: graphviz enables blueprint-graph visualization. Without it dimos logs
+# "graphviz not found, skipping blueprint graph" at startup (everything else still works).
+# sudo apt-get install -y graphviz
 
 # install uv
 curl -LsSf https://astral.sh/uv/install.sh | sh && export PATH="$HOME/.local/bin:$PATH"
+# NOTE: the `export` above only affects the current shell. Open a new terminal (or run
+# `source ~/.bashrc`) so `uv` is on PATH in later sessions — the installer also adds it there.
 ```
 
 ## Using DimOS as a library
diff --git a/docs/platforms/quadruped/go2/index.md b/docs/platforms/quadruped/go2/index.md
index 4e392f06ca..f02f46756e 100644
--- a/docs/platforms/quadruped/go2/index.md
+++ b/docs/platforms/quadruped/go2/index.md
@@ -27,7 +27,7 @@ uv pip install 'dimos[base,unitree]'
 
 ```bash
 # Replay a recorded Go2 navigation session
-# First run downloads ~2.4 GB of LiDAR/video data from LFS
+# First run downloads ~85 MB from LFS (extracts to a ~220 MB SQLite replay DB)
 dimos --replay run unitree-go2
 ```
 
@@ -110,6 +110,8 @@ From the command center ([localhost:7779](http://localhost:7779)):
 
 ```bash
 uv pip install 'dimos[base,unitree,sim]'
+# First sim run also downloads assets: it git-clones mujoco_menagerie and pulls a
+# ~60 MB mujoco_sim asset bundle from LFS (one-time; in addition to the replay data above).
 dimos --simulation run unitree-go2
 ```