The problem
Your team already does context engineering: RAG pipelines, project instructions, documentation wikis.1 Your AI tools have more organizational context than ever, and they still answer from the middle of their training data, because the knowledge that steers your organization was never written down.1 Why the architect made that call, how your senior people actually think, the methodology that makes your best people effective: it lives in their heads and in conversations that evaporate.2 When those people are busy, your team waits. (uncommitted) When they leave, that knowledge leaves with them. (uncommitted)
What we do
Aswritten turns your recorded conversations into a perspective: an individually owned, versioned data structure that installs into any AI and steers how it thinks.3 Every claim your AI makes from it traces to the human conversation it came from.4 Have an opinion, and here's why.4
How it works
Capture. Conversations are the input: meeting transcripts, AI sessions, voice memos.5 You open a chat with your own AI about what just happened, it drafts a memory, and your corrections become part of the record.5 A consultant-led discovery can bootstrap the perspective from an existing corpus of transcripts.6
Install. The perspective compiles into every AI tool your team uses via MCP: Claude, Claude Code, GitHub Copilot, Codex.7 Tool-agnostic by design.7 Cite answers where a claim came from.4 Introspect names the gaps the perspective has not filled yet.8
Deploy. The same perspective deploys as agents that answer in your name, with citations back to the people who decided, escalating to you in the channel where the conversation is happening when a question reaches past what the organization has settled.9
Three engagements
Discovery creates the perspective from an existing conversation corpus, and the report cites every finding back to its source.6 Onboarding transfers it: new humans and new AIs are the same recipient shape.10 Governance maintains and steers with it: the conversations that already steer your organization now also steer its AIs.11
In production
Aswritten runs in production today with a paying enterprise customer.12 The pilot began inside their architecture review process, bootstrapped from months of existing transcripts, and converted to a paid subscription within a month.12 Engineers reach the perspective through GitHub Copilot; product owners reach it through Claude desktop.12
Your data
The perspective lives in your git repository, version-controlled alongside your code.13 You own it completely.13 When your AI grounds a response, the perspective is rebuilt in memory from your own transactions, served, and released. (uncommitted) LLM calls run through your own API keys and our zero-data-retention proxy.13 On-prem, air-gapped, and sovereign deployments are available for regulated environments.13
Next step
A 60-minute call with Scarlet Dame, our founder. She will show you by doing it: interviewing you about one domain your team knows deeply, and demonstrating what your AI can do with that knowledge.
scarlet@aswritten.ai · aswritten.ai
In the March 19, 2026 strategy session, Scarlet defined a new category that moves beyond standard context engineering:
"aswritten defines a new category: curated expertise installation. It sits above context engineering — context engineering determines what AI knows; curated expertise installation determines how AI thinks."
This framing establishes that while traditional tools focus on raw information retrieval, true steering requires installing a specific methodology. The category name was later simplified when Scarlet dropped "curated" as redundant, cementing "expertise installation" as the core mechanism.
During the June 10, 2026 session, the team highlighted the acute pain of organizational knowledge bottlenecks, drawing on an industry advisor's raw feedback from earlier in the year:
"people have no fucking clue what to do when they join a company"
The advisor's follow-on, "just not having to stop what I'm doing to tell somebody something all the fucking time. Yeah, I'm in... I don't care what it costs," is the context-server pain in the buyer's own voice, already in the perspective. This pain of evaporating conversational context was later extended by Scarlet, who noted that these bottlenecks do not just limit individuals but paralyze entire recurring processes like the Architecture Review Board.
In the May 2026 pitch-reframe sessions, Scarlet established steering as the central organizing concept for the product:
"What a recurring decision body does — an Architecture Review Board, an investment committee, a law firm's partners' meeting, a clinical case conference — is steering a process of decisions: a group, or a person, recurringly deciding how the work should go."
This framework connects the high-level organizational use cases directly to the product's founding technical thesis, which treats narrative as a steering vector that guides how a language model behaves.
During the March 19, 2026 MVP redefinition, Scarlet identified the core trust mechanism of the product:
"The moment of value is when someone sees AI-generated text with citations showing where it came from. Extends from the hallucination pitch: you can trust where an AI is coming from the first time."
This focus on verifiable provenance was later validated independently by a senior technical evaluator during a June 25, 2026 demo call, where he praised the system's ability to return answers grounded in truth with clear dates and citations.
In an April 16, 2026 post-launch review, Scarlet realized that the product's core ingestion workflow was being underplayed in sales conversations:
"Both inputs — conversations with others (call transcripts) and conversations with herself (voice memos) — are downplayed in the pitch as 'this is what you do.'"
This realization was based on her own daily practice, where she relies on call transcripts and personal voice memos as the two primary inputs to continuously update her own perspective.
In her July 2, 2026 taxonomy session, Scarlet structured the enterprise lifecycle, defining the initial phase as a high-value entry point:
"Discovery — consultancy interviews → synthesized pitch plan, cited back to source, automatable with provenance"
This structured discovery process is supported by the technical capability to ingest large volumes of historical data. During a May 21 review, Scarlet confirmed that the onboarding workflow can confidently promise to turn six months of raw transcripts into a working perspective overnight using bulk AI summarization.
In a May 1, 2026 discussion, the customer's chief architect highlighted why the Model Context Protocol (MCP) is the ideal delivery mechanism for technical teams:
"For them, MCP is a productivity multiplier — a curated perspective they can pull into existing workflows rather than switching to a chatbot."
This integration allows developers to stay within their preferred environments, using a simple mode switch like the ", aswritten" suffix at the end of a prompt to trigger the installed perspective instead of generic model training data.
During the first week of the production pilot, the customer's chief architect heavily utilized the system's core capabilities, proving out the utility of the introspection tool:
"the heaviest user in week one: 69 calls, approximately 47% of attributed activity. He runs the full read-write loop himself: 29 remembers, 22 perspective loads, 12 introspects, 4 cites."
This active use of the introspect tool allows the system to identify missing context, which directly supports the pilot's goal of coaching team members by surfacing clarifying questions they need to answer before presenting to the Architecture Review Board.
In an April 24, 2026 session, Scarlet finalized the deployment architecture for external-facing perspectives:
"Unified Instrument: Scarce-Expert Filter with Goal-Gated Escalation"
This mechanism allows the perspective to act as an automated agent that represents the expert's worldview. It directly implements the broader product vision of goal-directed delegation with citable escalation, ensuring that when an AI agent encounters a question reaching past what has been settled, it routes the conversation back to the human decision-maker.
In her July 2, 2026 lifecycle taxonomy, Scarlet formalized onboarding as a core capability of the platform:
"Onboarding as a named, distinct use case: enabling an organization to onboard new staff because there's already an existing knowledge base. The recipient shape is the same whether the recipient is a human or an AI — both are being onboarded into an organizational perspective."
This perspective-driven onboarding model treats humans and AIs as identical targets for knowledge transfer. It sits adjacent to an earlier March 30, 2026 strategy claim where Scarlet suggested that organizational roles themselves should be represented as active perspectives contributed to by humans on a regular cadence.
During the May 26, 2026 positioning sessions, Scarlet carefully drew the line on how the product speaks to enterprise buyers:
"'Governance' names the customer's world — their meetings, their venues. Steering is the genus; governance is one species."
This distinction allows Aswritten to speak the language of enterprise decision-makers without getting misclassified as compliance software, positioning the platform instead as the active memory substrate that steers both human teams and their AIs.
The production pilot was established during an April 13, 2026 demo call, where the customer's chief architect identified a rapid path to value:
"a major client delivery iteration is the natural bounded pilot — it will generate heavy ARB traffic from a known group of people. The customer had ~6 months of ARB transcripts that could be ingested to bootstrap the perspective — a much faster path than formal extraction interviews."
By late June, the pilot had successfully transitioned into active production. A senior technical evaluator's demo record confirms that on June 25, 2026 the team was "really liking it so far," with engineers accessing the perspective through GitHub Copilot and product owners utilizing it via Claude desktop.
In a June 4, 2026 technical call, the architectural boundaries of the platform were firmly set to address enterprise compliance:
"aswritten is a processor, not a data store — enabling zero data retention for high-compliance environments"
This architecture ensures that the entire knowledge graph remains stored in the customer's own git repository. This zero-data-retention model directly resolves early adoption friction, such as the privacy concerns an early user raised in March regarding the visibility of sensitive or experimental work.