# The September 2026 Solopreneur Stack Update: Superhuman Docs, Model Controls, and RTX 5090 Sovereignty

> Audit your solopreneur stack this month. Learn how Superhuman Docs replaces Coda, Notion's new AI governance features, and why the RTX 5090 defines local inference.

- Source: https://solobrain-auto.nicheflash.com/blogs/september-2026-solopreneur-stack-update-superhuman-docs-model-controls
- Publisher: SoloBrain Automation
- Published: 2026-09-19
- Updated: 2026-09-19

- **Superhuman Docs replaces Coda:** Following the acquisition by Grammarly, Coda fully rebranded to "Superhuman Docs" in July 2026, adding general Model Context Protocol (MCP) access and a scaled data infrastructure.
- **Granular Model Governance:** Notion released a "Model Controls" feature (September 2026) that lets enterprise-level workspace owners dictate exactly which AI models their Agents are permitted to use.
- **The Consumer GPU Ceiling:** The NVIDIA RTX 5090 is now established as the definitive hardware choice for local LLM inference, natively running 70B-parameter models using 32GB of GDDR7 memory.

 ## What happened to Coda, and what is Superhuman Docs?

 If you were planning to build a new automation blueprint in Coda this week, you may have found yourself locked out or confused by the UI changes. The short answer is that **Coda no longer exists as a standalone product;** it has been fully integrated into what was formerly known simply as "Superhuman," the email client ecosystem of Grammarly.

 In October 2025, Grammarly announced it was becoming "Superhuman," and in December of that year, they finalized the acquisition of Coda. By July 2026, the entire operation officially migrated under the moniker **Superhuman Docs**. For solopreneurs, this is not merely a cosmetic rename—it represents a fundamental shift in how document-based databases handle external connections.

 Superhuman Docs introduces two critical capabilities that supersede the old Coda Packs:

 - **General Access to Model Context Protocol (MCP):** While previously restricted to beta developers, standard users now have permission to connect external tools directly to their internal data tables via MCP servers.
- **A New Data Infrastructure:** The rebrand comes alongside rebuilt backend logic designed for higher-scale operations, moving away from the smaller-dataset constraints that plagued many indie hackers.

 *The strategic implication:* If you were relying on Coda as a secondary data layer behind a Notion first brain, you must audit your automations immediately. Many legacy button triggers have moved to new endpoints within the Superhuman Docs API.

 ## How do Notion's new model controls affect my agents?

 As AI workloads move into our daily operational loops, costs can spiral unpredictably—especially when custom agents are left unsupervised on heavy computation tasks. In the September 9, 2026 release of Notion 3.7, the platform addressed this anxiety head-on with a feature called **Model Controls**.

 This update allows workspace owners on Business and Enterprise plans to strictly curate the roster of Large Language Models (LLMs) that Personal Agents and Custom Agents are permitted to utilize. You can assign a default model for heavy drafting tasks while reserving highly expensive reasoning models (such as OpenAI o-series or Anthropic Claude Opus variants) strictly for manual user queries.

 This effectively shifts the responsibility of budget management from the freelancer doing the math back to the system architecture itself. Furthermore, combined with the **AI Meeting Notes** feature launched in August—which now automatically triggers Custom Agents upon transcription—these controls ensure that passive content repurposing pipelines don't inadvertently consume your monthly compute allowance.

 ## Is the RTX 5090 finally viable for a desktop second brain?

 For years, building a true "sovereign second brain" locally required spending thousands on professional-grade GPUs like the A100 or H100. That barrier has recently collapsed due to the mass availability of the **NVIDIA GeForce RTX 5090**.

 Released late in 2025 but dominating the discourse into 2026, the RTX 5090 offers 32GB of GDDR7 memory with bandwidth reaching approximately 1,792 GB/s. Benchmarks conducted by the open-source community show this card can successfully run high-fidelity 70B-parameter models—such as DeepSeek-V3 or GLM-5 quantized to 4-bit precision—directly on a single consumer workstation.

 | GPU Tier | VRAM Capacity | Viable Max Parameter Size (Q4) | Best Use Case |
| --- | --- | --- | --- |
| RTX 4090 | 24 GB | ~34 Billion | Fast summarization & text rewriting |
| **RTX 5090** | **32 GB (GDDR7)** | **~60 - 72 Billion** | **Sovereign knowledge retrieval & local agents** |
| A100 | 40 GB+ (PCIe) | >100 Billion | High-throughput commercial training |

 The speed difference is stark; the RTX 5090 generates text roughly 35% to 50% faster than its predecessor, the RTX 4090. For a solopreneur wanting to run a semantic tagging system or a local multimodal pipeline without sending data to the cloud, this is currently the most cost-effective entry point into total privacy sovereignty.