ASG · AGENTIC ENGINEERING, DELIVERED ON bolde.

Make AI do real work inside the business you already run.

Bring us one workflow that matters. ASG’s Deployment Engineers connect the systems it touches, configure our product bolde. around the work, and take responsibility for the agreed scope until it is working in production.

For operating, finance and legal leaders with one workflow that crosses several systems and has to be right.

Request a Briefing

45 minutes. No fee. Decide whether defining a deployment scope together makes sense.

bolde. whole, by part, or not at all — Phase 0 decides.

45 MINUTES JASON OR MANBIR PRINCIPAL-LED BRIEFING

Illustrative. Sources stay linked to the report they fill.

01 — OUTCOME

02 · AN ILLUSTRATIVE DEPLOYMENT

What a deployment looks like.

A finance team rebuilds a quarterly investor report by hand from four systems. Here is the same work as a living report in a shared workspace: it stays linked to its sources, changes arrive as suggestions, and the owner approves.

Example

Keep one report current, from its sources to sign-off.

Illustrative workflow: a finance team rebuilds its quarterly investor report by hand from four systems. Here it is as a living report in one shared workspace: figures stay linked to approved sources, a source change arrives as a suggestion rather than an overwrite, and the owner approves before anything issues. Below is the acceptance sheet the team agrees before it goes live.

Illustrative workflow · scripted example, not a live bolde. session or client

See the complete workflow

What you run today

Fund accounting system
Document store
CRM
Email

What the workflow does

The work, after

Quarterly summary Example review
Report item	Review state
Portfolio figure	Linked to its source record
Investor commentary	Edited by the analyst
Source refresh	Suggested update · awaiting review
Owner approval	Recorded in activity history

The report owner decides whether it is ready to issue.

Example input: Finance records from four approved systems
Expected output: A living quarterly investor report with cited figures, kept current each quarter
Scope: Uses Lake and Space only; connects to the client’s existing model route.
Review owner: Customer report owner

Example review rule: when a source changes, propose the update to a person; never overwrite their edit, and never issue without the owner’s approval.

4 of 4 example checks holding

Each report figure links to its approved source record.
People edit the report directly; each commentary line keeps its source link.
A source change arrives as a suggestion and does not overwrite a person’s edit.
An agent suggestion counts only after a person approves it, and the approval is recorded.

Not product UI, client evidence, signed acceptance criteria, or an operative SOW. The customer and ASG agree actual criteria before the Build.

Illustrative. Sample workflow, not a client.Fund accounting, documents, CRM and email feed four stages: connect approved sources, draft the living report, refresh as sources change and approve in the shared workspace. Each stage fills one row of the report at right and is held to one check on the acceptance sheet below it.

Who is in the workspace

Report ownerOwner · grants access, approves
Finance analystCan edit the report
Investor relations leadCan comment on commentary
Drafting agentSuggests only · approved sources only

Belonging to the workspace does not make the report public. The agent sees only the sources and items approved for it.

Activity history records who joined, access changes, edits, suggestions and approvals.

Living documents, slides and spreadsheets keep their figures and citations current as approved sources change, without overwriting your edits.Planned — claim-20, open in the Trust Center
Details
Owner
CTO
Status
Planned
Required evidence
A product release record for bolde. Evergreen. bolde.ai/features lists Evergreen as a planned release target within one month (snapshot 2026-10-21).
Not confirmed for your deployment.
A shared workspace brings people, agents and work together; owners set view or edit access, and an agent suggestion becomes an edit only when a person approves it.On request — claim-21, open in the Trust Center
Details
Owner
CTO
Status
On request
Required evidence
A dated release or deployment record for bolde. Space. bolde.ai/features lists Space as Phase I, in progress or shipping now (snapshot 2026-10-21); no record has been verified for this site.
Not confirmed for your deployment.

The pattern follows how bolde.ai describes bolde. Evergreen and bolde. Space. Status for each is on bolde.ai.

Which bolde. components are involved?

Component release status is on bolde.ai.

Make it fail

Break the example on purpose.

Each failure stops the report in a stated place, with a named person deciding what happens next. There is no score.

Remove a source
Unsupported in this example
Cannot establish
Where the commentary came from, once the document store is disconnected.
Who decides
The report owner.
What permits progress
Reconnect the source, or take the unsupported lines out of the report.
Records disagree
Unresolved: owner decides which record governs
Cannot establish
Which of two conflicting records is correct for the portfolio figure.
Who decides
The report owner, under the rule agreed in configuration.
What permits progress
The owner records which source governs; the row is checked again.
Source changes after an edit
Suggested, not applied: the edit stands
Cannot establish
Whether the analyst’s wording or the refreshed source should govern the commentary.
Who decides
The analyst who made the edit, then the report owner.
What permits progress
A person accepts or declines the suggested update; both choices go into the activity history.
Agent reaches past its scope
Blocked: outside the approved sources
Cannot establish
Anything from a source the drafting agent was not approved to read.
Who decides
The report owner, who sets what the agent may see.
What permits progress
The owner approves the source for the agent, or the agent works without it.
Withhold approval
Returns to Build; decision kept
Cannot establish
That the report is ready to issue.
Who decides
The report owner. Nothing issues on their behalf.
What permits progress
The decision stays on the record and the work returns to Build until the criteria are met.

Write your own criteria

03 · HOW FAR YOU GO

Keep what you run. Choose how far bolde. goes.

Your ERP, CRM, document system, data warehouse and cloud stay where they are. You decide how much of your work moves into bolde., and you decide later whether to go further.

Discuss the smallest scope that would be useful.

Inside your tools

Agents work in the email, documents and systems your team already opens. You decide which actions require a record.

Agent and connection options · On request

A shared workspace

Your team asks, reviews and approves in one place, with links back to source records where configured.

Space · On requestConsole with Operator · Planned

The full stack

Agents, workspace, models, data controls and hosting, configured to your security level.

Model and hosting options · On request

04 · WHO DOES THE WORK

The people you brief lead the deployment.

CEO · CO-FOUNDER

Jason Wareham

Answers for how each deployment is governed: scope, security level, and the contract terms your counsel will review.

CTO · CO-FOUNDER

Manbir Gulati

Leads the bolde. platform and the Deployment Engineers who connect it to your systems.

THE TEAM ON YOUR DEPLOYMENT

Deployment Engineers

Connect bolde. to each of your systems, ship it to production, and keep the connections they own working after go-live, as set in your proposal.

05 · THE PATH INTO PRODUCTION

From the first conversation to a workflow your team runs.

01
Briefing
Forty-five minutes with a principal. Bring one workflow and one constraint. If bolde. is not the right fit, we say so in the meeting.
02
Configuration
Phase 0 of your bolde. deployment. We configure bolde. for the workflow and your security level, and write down the scope and what acceptance means.
03
Build
Deployment Engineers connect bolde. to your systems and build the agents, against the acceptance criteria agreed in configuration.
04
Acceptance
Your team tests the workflow on your own examples. It goes live when it meets the criteria.
05
Operation
We run and improve bolde. for you. Each new workflow is added to the same agreement.

Your agreement is designed to cover configuration, build and subscription, with your security level in a written schedule. Planned

06 · WHERE IT RUNS

Choose the models. Choose where bolde. runs. Set the security level in writing.

bolde. is our platform, and we use as much of it as the work needs — all of it, one part such as Lake, or none. Start with one workflow, then expand.

Model routing is independent: bolde. connects to the models you choose and runs at the security level you set.

All of bolde.
One part, such as Lake
Your models, any level

Parts of bolde., such as Lake, can be deployed on their own inside your existing stack. On request — scoped and confirmed for your deployment.

Model route, hosting and access are three separate decisions, confirmed per deployment.

We work with the model route your policy calls for: open-source and open-weight models run privately, frontier model APIs under zero-data-retention terms, or full frontier APIs where your policy allows. Each route is set by agreement. On request

Customer-provided models are also possible by agreement. The exact model is confirmed for each deployment.

Four security levels

01OpenPublic frontier models, by separate agreementOn request
02HybridPrivate models for sensitive work; public frontier models for the rest, by separate agreementOn request
03PrivatePrivate models onlyOn request
04SovereignPrivate or customer-provided models on customer-controlled hardwareOn request

Ask three questions: what data is involved, where processing happens, and who has access.

Compare the four levels →

We publish the status of every material claim, including the ones not yet verified (0 of 21 verified). Open the Trust Center

FIELD NOTES

Field Notes · Read before you brief

FIELD NOTE 01 · SECURITYFour security patterns for AI, and how to pick yours.Before you choose a model, decide what each model may see. Four security patterns for AI, from Open to Sovereign, and how to pick yours.

STRAIGHT ANSWERS

Straight answers to what buyers ask first.

How is it priced?

We agree the scope, fees, and subscription terms before work begins. Bring us a workflow and we can prepare a proposal for your review.

Where does my data go?

Before deployment, we document the proposed model route, hosting, access path, data classes, and return terms for your review. The agreed boundaries go into your deployment agreement.

Can we choose the model and still choose where it runs?

Yes. Model and hosting choices are separate. We can assess public frontier models by separate agreement, private open-source or open-weight models where supported, and customer-provided routes by agreement. The exact model, data classes and host are confirmed for your deployment.

Do we have to replace our systems?

We start by assessing connections to the systems you already run. Any required changes are identified in scope.

How long does it take?

It depends on the systems involved. In the Briefing we tell you what drives the timeline for your workflow. During configuration (Phase 0 of your bolde. deployment) we agree scope, timeline and target go-live date in writing before the Build.

Do we have to use bolde.?

No. Phase 0 decides the right answer for the work. Most often that is bolde., whole or in part. When it is not, we say so, and we build or recommend what is.

What happens if we leave?

Your data is yours. You get it back when you leave. The return format and period are set in your agreement.

Do you do work without bolde.?

We only take on work we deliver on bolde.. If you need something else, we’ll say so in the Briefing.

Questions buyers hesitate to ask, answered with status →