name

sima-environment-setup

description

Use when setting up a SiMa.ai development environment from scratch — installing sima-cli, authenticating to the SiMa Developer Portal, installing the Palette Neat SDK and Model Compiler, pairing a Modalix DevKit, compiling a model to an MPK package, and building or running NEAT C++/Python and GenAI/LLiMa applications. Do not use for Neat Library application design or API selection (use neat-application-builder), for hardware bring-up such as serial, network, or firmware procedures (follow the hardware docs), or for repository maintenance.

SiMa Environment Setup

Overview

Take a host (and optionally a Modalix DevKit) from zero to a working SiMa.ai development environment. The single source of truth is the published agent onboarding page — fetch it first and follow its numbered phases; it carries the exact commands and the currently supported version pins:

Raw (preferred): https://developer.sima.ai/agents.md or https://developer.sima.ai/llms.txt
Rendered: https://developer.sima.ai/agents
Local fallback when working inside the docs repo: src/pages/agents.md (note: the source contains unsubstituted %key% version tokens; values live in src/versions.json)

This skill encodes the workflow and guardrails; the page carries the commands.

Workflow

Gather inputs from the human before running anything: SiMa Developer Portal credentials (community.sima.ai), the DevKit IP address (if a DevKit is in scope), a Hugging Face token (GenAI only), and the target silicon (default modalix).
Fetch the source of truth: curl -fsSL https://developer.sima.ai/agents.md.
Pick the phase subset from the page's decision matrix (Phase 0) that matches the stated goal. Do not run phases the goal does not need.
Execute the phases in order. Stop at every Verify block, run it, and confirm the expected result before continuing. Never skip verification.
sima-cli login and the SDK installer are interactive — hand those prompts to the human rather than trying to automate credentials.
On any failure, consult references/troubleshooting.md, then the page's Phase 9. Retry the failed Verify before moving on.
Before claiming success, run the end-state checks in references/verification.md for the phases you executed.
To route follow-up questions ("where do I read about X?"), use references/documentation-map.md.

Defaults

Target silicon modalix (gen2); mlsoc/davinci (gen1) only when the human says the hardware is the legacy generation.
Primary host assumption: Ubuntu 22.04/24.04 x86_64 with Docker Engine.
Install the SDK with sima-cli neat install sdk@<channel> using the channel named on the page; do not pin versions from memory.
Everything installs through sima-cli — do not docker pull or pip install SiMa components directly unless the page says so.

Boundaries

Do not design application code or choose between Model/Graph/GenAI APIs — defer to the neat-application-builder skill and the Develop Apps docs.
Do not perform hardware bring-up (serial cabling, network configuration, firmware flashing) beyond linking the human to the hardware pages the onboarding page references.
Do not invent commands, package names, or version numbers from memory — always re-fetch the raw page.
Do not claim success without the Verify results to show for it.

References

references/verification.md
references/troubleshooting.md
references/documentation-map.md

SiMa Environment Setup

Overview

Raw (preferred): https://developer.sima.ai/agents.md or https://developer.sima.ai/llms.txt

Rendered: https://developer.sima.ai/agents

Local fallback when working inside the docs repo: src/pages/agents.md (note: the source contains unsubstituted %key% version tokens; values live in src/versions.json)

This skill encodes the workflow and guardrails; the page carries the commands.

Workflow

Gather inputs from the human before running anything: SiMa Developer Portal credentials (community.sima.ai), the DevKit IP address (if a DevKit is in scope), a Hugging Face token (GenAI only), and the target silicon (default modalix).

Fetch the source of truth: curl -fsSL https://developer.sima.ai/agents.md.

Pick the phase subset from the page's decision matrix (Phase 0) that matches the stated goal. Do not run phases the goal does not need.

Execute the phases in order. Stop at every Verify block, run it, and confirm the expected result before continuing. Never skip verification.

sima-cli login and the SDK installer are interactive — hand those prompts to the human rather than trying to automate credentials.

On any failure, consult references/troubleshooting.md, then the page's Phase 9. Retry the failed Verify before moving on.

Before claiming success, run the end-state checks in references/verification.md for the phases you executed.

To route follow-up questions ("where do I read about X?"), use references/documentation-map.md.

Defaults

Target silicon modalix (gen2); mlsoc/davinci (gen1) only when the human says the hardware is the legacy generation.

Primary host assumption: Ubuntu 22.04/24.04 x86_64 with Docker Engine.

Install the SDK with sima-cli neat install sdk@<channel> using the channel named on the page; do not pin versions from memory.

Everything installs through sima-cli — do not docker pull or pip install SiMa components directly unless the page says so.

Boundaries

Do not design application code or choose between Model/Graph/GenAI APIs — defer to the neat-application-builder skill and the Develop Apps docs.

Do not perform hardware bring-up (serial cabling, network configuration, firmware flashing) beyond linking the human to the hardware pages the onboarding page references.

Do not invent commands, package names, or version numbers from memory — always re-fetch the raw page.

Do not claim success without the Verify results to show for it.

References

references/verification.md

references/troubleshooting.md

references/documentation-map.md