Agent Docs

Development guides for AI agents working on the Rendered.ai platform.

The material is split by audience. Pick the section that matches what you're building.

Agentically developing services (ML pipelines, data processors, API workers) that run inside a Rendered.ai Agent Studio workspace and interact with platform resources (datasets, volumes, models, channels). Applies to first- and third-party service developers.

• skills/agent-studio — Agent Studio platform reference: workspace storage layout (local NVMe vs FUSE), the FUSE timeout rule for long-running jobs, on-demand (test_service, run_service) vs persistent (run_persistent_service) execution, the renderedai CLI and anatools SDK, the MCP raiservices-local toolchain, and layered platform rules. Packaged as an installable skill; also readable directly as skills/agent-studio/AGENT_STUDIO.md.

Authoring synthetic-data channels — Blender, DIRSIG, Omniverse, and the graph/node system they run on.

• AGENT.md — General channel development: node/package/Dockerfile anatomy, the local ana dev container and path mappings, node class structure and anatools object types, package volumes, channel config, determinism, annotations, deployment workflow, and common mistakes.
• AGENT_BLENDER.md — Blender-specific patterns: scene management, materials, lighting, cameras, and rendering within Rendered.ai channels.
• AGENT_DIRSIG.md — DIRSIG-specific patterns: the dirfm driver library, glist object/instance model, motion/flex-motion engines, platform sensors, truth-band annotations, and the DIRSIG5 simulation lifecycle.
• AGENT_OMNIVERSE.md — Omniverse / Replicator-specific patterns: the declarative-graph mental model, randomizer contract, custom writer + annotators, preview mode, determinism with rep.set_global_seed, and stage-state leaks.
• AGENT_GRAPH.md — Graph YAML authoring: node anatomy, hash versioning, port definitions, link rules, YAML 1.1 pitfalls, common errors, and graph-editing checklist.
• AGENT_SDK.md — anatools Python SDK and companion CLIs (anamount for dataset mounting, anatools-download-dataset for bulk download): authentication, graph upload, dataset creation, log download, and platform workflows.

Narrow, pitfall-focused agent-skill packages under skills/ — each pairs a trigger description with a prescriptive reference for one recurring, high-friction task:

• skills/channel-node-io — Silent-failure patterns in a node's exec(): unindexed self.inputs, output-key mismatches, the unlinked-optional-input "" sentinel, and FileObject/DirectoryObject attribute access.
• skills/channel-schema-validation — Node schema YAML pitfalls that only surface as a red border in the web editor (type: float, oneOf branch collisions, default/branch mismatches, the literal-or-wired-node port pattern).
• skills/channel-replay-run — Reproducing a platform dataset run locally with ana: dataset-name→ID lookup, seed/interp_num extraction, graph.json-vs-local-YAML drift, and pulling per-run platform logs.
• skills/blender-blend-inspect — Headless bpy inspection of a .blend file inside a channel's Docker image (object positions, actions, cameras, Geometry Nodes inputs).

Include these files in your channel repository (e.g., at the repo root) so AI coding assistants automatically pick them up as context. They are designed to reduce common mistakes and accelerate development of new nodes, graphs, and channels.

Update the guides as you discover new patterns or pitfalls. Keep entries concise and example-driven.