The provenance database for accountable AI work

AI-generated work needs a chain of custody.

checkpt saves the artifact and the working record together, so organizations can trust AI outputs and build with shared context.

Why now

AI work is already flooding the enterprise. Strategy is not.

Teams are using Claude, Codex, Cursor, Copilot, internal agents, and frontier-model tools in parallel. The output is real work product, but the working record is scattered, private, or gone.

No strategyEvery team experiments alone, so the organization cannot see what works.

No recordPrompts, evidence, model choices, approvals, and decisions disappear.

No shared intelligenceEach platform builds its own memory while the enterprise loses the bigger pattern.

A neutral intelligence plane brings order before agent sprawl hardens into operating chaos.

The accountability gap

Companies are producing work they cannot prove.

Across code, specs, analyses, proposals, decks, customer deliverables, and decisions, teams keep the final artifact but lose the causal chain that produced it.

The hard questions

Which inputs were used? What did the agent see? Which evidence mattered? What changed? Who approved it? Can we explain, reproduce, continue, or audit the work?

When the record is missing

Audits fail. Decisions become oral history. Teams repeat experiments. Quality depends on whoever remembers the run.

checkpt

Save the artifact and the working record together.

checkpt captures the inputs, context, conversation, evidence, output, verification, and lineage behind agent-generated work, then packages them into portable checkpoint records.

Input→Context→Conversation→Evidence→Artifact→Lineage→Intelligence

inspectableauditablereusableportablemodel-neutralteam-visible

Technical wedge

Everything is markdown. So the database should be markdown-native.

Markdown is the shared language of AI toolchains, agents, docs, prompts, plans, specs, and code-adjacent work. checkpt turns that substrate into a provenance database.

Checkpoint capsule

checkpoint.md
the human-readable work record

manifest.json
structured metadata, lineage, environment, and policy hooks

transcripts · evidence · verification
the materials needed to review, continue, and improve the work

New data architecture

Intelligence should live inside the record, not be extracted later.

Today

Store the output. Lose the run. Scrape logs later. Reconstruct context from Slack, docs, repo history, and memory. Analytics arrives after the causal chain is already broken.

checkpt

Capture context, evidence, evaluations, and lineage at creation time. The work record is already structured enough to compare strategies, improve workflows, and train better organizational habits.

Capture→Evaluate→Compare→Reuse→Optimize→Future product layer

Software first

Source control proved that accountable work history creates enormous value.

GitHub and GitLab show the scale of developer workflow infrastructure. But Git tracks snapshots and diffs; agent-generated work also needs intent, prompts, evidence, evaluations, and lineage.

180M+ GitHub developersGitHub Octoverse 2025

630M projects on GitHubMassive surface area for work history

$759M GitLab FY25 revenueDeveloper infrastructure budgets are real

Sources: GitHub Octoverse 2025, GitLab FY25 results.

Cross-platform moat

Neutral records beat platform-owned memory.

Enterprises will not standardize on one model, one agent, one editor, one repo, or one document system. checkpt gives them a single intelligence plane across the messy reality.

Models Claude · OpenAI · Gemini · local

Tools Cursor · Codex · Copilot · agents

Artifacts code · docs · decks · decisions

Environments local · cloud · regulated workspaces

Order from chaosIf every frontier platform DIYs its own memory, corporate intelligence fractures. checkpt keeps the record portable, inspectable, and owned by the enterprise.

Platform upside

Accountability compounds into corporate intelligence.

Once the working record exists, checkpt can help organizations see which agents, prompts, evidence, workflows, teams, and strategies produce the best outcomes.

Accountability layerWho or what produced the artifact, from which context, under which constraints.

Operational layerContinue from the right state, reuse strong workflows, reduce duplicated experimentation.

Intelligence layerReporting, pattern detection, evaluation, optimization, and future product capabilities.

The wedge is provenance. The company becomes the intelligence layer for agent-generated work. GTM starts with open-source developer pull, then regulated design partners, then an enterprise control plane for identity, policy, sync, audit exports, evaluation, and reporting.

Why Wes

The right person at the right time.

checkpt comes from a lifelong builder who has lived inside messy, regulated systems and has the impatience to call out status quo tooling when it stops matching reality.

Physicist First-principles reasoning about complex systems and uncertainty.

Healthtech CTO Regulated systems where auditability and trust are not optional.

Maverick Mavericks are visionaries who operate on their own terms. They break molds, challenge assumptions, and lead through bold action.

Maverick framing adapted from The Predictive Index Maverick Reference Profile.

Founder-market fit

A physicist’s systems instincts, healthtech scar tissue, and daily agent use converge on the same gap: AI work needs a durable record.

Pre-seed SAFE

Raising $500K to reach Seed-quality signal in 6 to 9 months.

Pilots: launch larger design-partner pilots in regulated and enterprise environments.
Adoption: drive developer usage through open source, examples, agent integrations, and skills.
Team: add engineering capacity to develop the database, provenance records, and governed workspaces.
Seed signal: prove repeatable usage, enterprise pull, and the intelligence layer narrative.

Seed signal

What this buys: enough product depth, pilot evidence, and adoption signal to make the Seed round about acceleration, not validation.

Thank you

Your time is valuable.

The deck is the start of the conversation. The question now is how quickly accountable AI work becomes enterprise infrastructure.

Why not just RAG?

RAG retrieves context. checkpt creates accountable context.

Retrieval can help an agent find relevant knowledge. It does not prove what the agent actually saw, used, decided, changed, evaluated, or shipped.

Database plus RAG

Store artifacts, logs, docs, and chats. Search them later. Hope the right fragments reconstruct the work well enough to answer the next question.

checkpt

Create the work record while the work happens. Inputs, evidence, transcripts, decisions, outputs, evaluations, and lineage are bundled at the source.

RAG is an access pattern→checkpt is the system of record→better source data makes better retrieval

How lanes create leverage

A checkpoint can carry context into a new lane of work.

Teams do not just save what was made. They bundle the context that made it possible, then share that bundle with another agent, team, or workflow to create the next artifact from the right state.

Original laneCapture the artifact, prompts, evidence, transcript, verification, and decisions.

Checkpoint capsulePackage the working record as markdown-native provenance with structured metadata.

Shared contextMove only the needed context into another controlled lane, without copying platform memory.

New workCreate code, docs, plans, analyses, or deliverables with lineage back to the source.

Why it mattersContext becomes portable working capital. Each lane can move forward with accountability, reuse, evaluation, and a cleaner path back to the decisions that shaped the work.

Links and public source

Website, open-source repos, and current build surface.

Public links for the investor follow-up, plus the local components that show the product surface built around the markdown-native provenance database.

Website

checkpt.co
Product home and future hosted workspace surface.

Open-source repos

checkpt/cli
CLI, agent skills, Codex wrapper, checkpoint workflow.

checkpt/core
Shared checkpoint primitives and provenance model.

checkpt/daemon
Local capture daemon for appending records and creating checkpoints.

ckptdbLocal markdown-native database.

WebsiteSignup, authentication, and administration.

Agent NativeSkills and plugins for Claude, Codex, and Cursor.