Claims keep their source, confidence, and visibility.
Your work. Your voice. Your evidence.
Marrow is the private identity layer for personal AI agents: a living corpus of real work, real evidence, and real voice.
What changes
Renderers and agents read from the same private substrate.
Corrections become inspectable records, not hidden state.
Style is grounded in writing samples and accepted edits.
Identity, genesis, knowledge, core, connectivity
One person's professional substance, made queryable.
Identity
The stable professional facts a person wants agents to carry: roles, projects, skills, credits, publications, credentials, and the claims that connect them.
Genesis
Every claim keeps a line back to where it came from: a source note, thesis, repo, article, accepted correction, or public record.
Knowledge
Marrow stores what the person knows through the work that proves it, not through a flat list of keywords or self-descriptions.
Core
A private graph of notes, claims, entities, projects, edges, facets, provenance, confidence, and voice evidence.
Connectivity
Agents, portfolios, CV tools, email surfaces, and future personal assistants can draw from the same source-grounded substrate.
The context problem
Personal agents are only as good as the context they carry.
Chat history is noisy. Resumes are shallow. Files are scattered. Prompt memory is hidden and hard to inspect.
That is how generic AI ends up sounding polished while losing the work, judgment, and voice that make a person specific.
From corpus to context
A source-grounded path from real work to agent action.
-
01
Ingest the corpus
Writing, work history, project notes, PDFs, portfolios, and public references.
-
02
Build the graph
Notes, claims, entities, projects, edges, facets, provenance, confidence, and embeddings.
-
03
Retrieve context
Agents ask what matters and get relevant identity, evidence, and voice records.
-
04
Generate with evidence
Outputs can show the claims and style evidence used instead of hiding the source.
-
05
Learn from correction
Accepted edits become inspectable updates, not hidden prompt memory.
Product proof
From source material to evidence-backed representation.
Professional target
Private corpus
Cited context
Accepted output
A high-stakes professional prompt goes in. Marrow retrieves the relevant work, source records, public and private evidence, style observations, and caveats.
The output can show why each claim is present, which evidence supports it, what was omitted, and how an accepted correction would update the corpus.
View the proof pathReal work, real evidence, real voice
The human signal is preserved as structure.
Work
Projects, roles, artifacts, outcomes, thesis work, publications, credits, and source material.
Evidence
Claims linked to source notes, public references, confidence, visibility, and provenance.
Voice
Writing samples, style observations, vocabulary, anti-patterns, and accepted edits.
Trust boundary
Verifiable does not mean public.
Marrow distinguishes public evidence, private evidence, inferred evidence, and directly confirmed evidence.
The person controls what gets exposed, to whom, and why. Agent feedback enters as source-backed records, corrections, or superseded claims that can be inspected.
Built for the personal agent ecosystem
Marrow is the substrate other systems speak from.
Tailor
Generate professional profiles from source-backed identity records.
Portfolio
Keep public pages current without losing provenance.
Writing agents
Draft in the person's register using style evidence.
Personal agents
Answer capability questions with evidence and caveats.
Future tools
Propose new claims, corrections, and source links.
Design partner alpha
For people whose professional identity is too nuanced for generic AI.
Built for deep work, public evidence, original writing, research, projects, founder material, and agent workflows where accuracy is part of the product.
Request a design partner slot