Sombra
Sign in
8 articlesShared 2 weeks agoLive

Sombra / About

What Sombra is and why it exists — product overview, core features (URL saving, collections, context distillation, MCP integration), and the research context: cognitive offloading, PKM, and the rise of context engineering.

Back to collection

The Research Context: Why Sombra Exists

2 weeks ago

The Research Context: Why Sombra Exists

Sombra sits at the intersection of three converging ideas: cognitive offloading, personal knowledge management, and the emergence of AI context engineering as a discipline. Understanding these gives you the "why" behind its design choices.

1. Cognitive offloading: extending the mind into the world

The idea that humans use the environment as a cognitive extension is not new. Andy Clark and David Chalmers' 1998 paper The Extended Mind articulated the philosophical case: cognition doesn't stop at the skull. When you write a shopping list, you've offloaded the memory task to the world. The list and your brain form a single cognitive system.

Risko and Gilbert (2016) formalised this as cognitive offloading — the act of reducing mental processing requirements by externalising information. Research consistently shows it improves task performance, especially at high cognitive load. Writing something down doesn't just remind you later; it frees working memory to think about something harder right now.

The implication for developer tooling is direct: the more your tools handle retention and retrieval, the more cognitive bandwidth you have for the reasoning that actually matters.

Sombra is a cognitive offload device for your research. Save now, think later — or let your AI agent think instead.

References:

  • Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19.
  • Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688.

2. Personal knowledge management: the problem of structured recall

Personal knowledge management (PKM) as a discipline traces to Davenport (1998) and the 1999 working paper by Frand & Hixon. It's the practice of collecting, classifying, storing, searching, retrieving, and sharing knowledge in daily work — a bottom-up approach to making yourself and your team smarter over time.

PKM has historically focused on human retrieval: how does a person find what they saved? The tool landscape reflects this — Notion, Obsidian, Roam, Logseq are all fundamentally human-facing recall systems. They optimise for browsability, backlinks, graph traversal, and visual organisation.

The field is undergoing a shift. As AI agents become central to how knowledge workers operate, the question changes: how does an agent retrieve what's been saved? The interface is no longer a browser sidebar or a graph view — it's a token window. The retrieval mechanism isn't search or hyperlinks — it's context injection.

Sombra is a PKM tool redesigned for this new retrieval paradigm. Collections aren't organised for you to browse; they're organised so an agent can receive them as coherent context. Distillation isn't a summary for future-you; it's a compressed signal optimised for an attention budget.

3. Context engineering: the discipline that makes agents work

In 2025, "context engineering" emerged as the central discipline in AI agent development — the practice of designing systems that provide the right information to a model, in the right format, at the right time.

Anthropic's engineering team defines it as "the discipline of designing and building dynamic systems that provides the right information and tools, in the right format, at the right time." Simon Willison and others popularised the framing as the natural successor to prompt engineering: where prompt engineering focused on instruction quality, context engineering manages the entire information environment an agent operates within.

The core insight: most agent failures are context failures. Models are capable — but they fail when given too much irrelevant information, when context is structured poorly, or when the right information simply isn't present. Research shows performance degrades sharply with context overload (meaningful degradation often begins well under 100k tokens), and that a 39% average performance drop occurs when prompts are fragmented inconsistently.

The implication for a knowledge tool: raw saved pages are poor context. 40 pages of documentation injected wholesale into an agent's context window is actively harmful — it introduces noise, triggers distraction failure modes, and exhausts the model's attention budget on content that doesn't affect the task.

The distillation layer in Sombra exists precisely because of this. A well-written collection context is dense, structured, and signal-only — the kind of context engineering that makes agents perform rather than hallucinate.

References:

4. MCP: the infrastructure that makes it all work

The Model Context Protocol was open-sourced by Anthropic in November 2024 as a universal standard for connecting AI systems to external tools and data sources. Adopted by OpenAI, Google DeepMind, Microsoft, GitHub, AWS, and Bloomberg within its first year, it became the standard infrastructure of the agentic AI ecosystem — donated to the Linux Foundation's Agentic AI Foundation in December 2025.

Before MCP, connecting an AI agent to your data required custom integrations for every tool and every client. MCP solves what Anthropic called the "N×M problem" — you build one MCP server, and any MCP-compatible client can use it. It's the USB-C standardisation moment for AI context.

Sombra's MCP server means your entire research library — every saved page, every collection, every distilled context note — is accessible to any AI tool you use, without any manual copying, file management, or re-configuration. Connect once. Every agent you use benefits immediately.

References: