OakCred Agents

Private AI agents, customised for your company

Private AI agents
for operations.finance.sales.legal.support.your company.

We design custom AI agents around your workflows and your data, then deploy them on your own servers. Nothing you connect is ever sent to an AI company: not to us, and not to anyone else.

Runs on your servers Never used for training You own the code
Start a build Browse agents

Runs on infrastructure you already trust

Amazon Web ServicesAmazon BedrockGoogle WorkspaceKubernetesTerraformPostgreSQLOpen-weight LLMs

Approach01

Off-the-shelf AI tools ask you to send your company's data to someone else's model. We build the opposite: agents shaped around how your company works, running entirely inside it.

1a

Built around your work.

Every agent is designed for a real workflow in your company, such as reconciling invoices, reviewing contracts or briefing a team. It isn't a generic chatbot.

1b

Grounded in your data.

Agents connect to Google Workspace, your CRM, databases and internal tools through one private data layer, so they answer with your context.

1c

Nothing leaves your walls.

We deploy everything on your servers. Your data never reaches an AI company, and there's no connection back to us after handover.

1d — The private data layer

Hover a tool or an agent to trace its path

Agent Catalogue02

Start from a proven agent design, or bring us any workflow. We customise each one to your processes, your tools and your data, then deploy it privately alongside the rest.

On demand
Knowledge

Company Knowledge Agent

Answers questions about projects, clients and decisions with citations, drawing on everything your teams have written.

GmailDriveDocs
Continuous
Operations

Workflow Automation Agent

Watches inboxes, forms and sheets, then plans and carries out multi-step work. Sensitive steps wait for a person to approve them.

GmailSheetsForms
Continuous
Operations

Meeting-to-Action Agent

Turns meeting transcripts into decisions, owners and due dates, then follows up until the work is done.

MeetCalendarChat
On demand
Sales & Support

Sales Research & Proposal Agent

Researches an account from your CRM and past deals, then drafts proposals in your company's voice and pricing structure.

CRMDriveGmail
Continuous
Sales & Support

Customer Support Agent

Drafts replies from your docs and past tickets, sends routine answers on its own and escalates edge cases to your team.

HelpdeskDocsGmail
Continuous
Finance & Data

Invoice Reconciliation Agent

Matches invoices arriving in Gmail against purchase orders and your ledger, and flags mismatches before they're paid.

GmailSheetsERP
Scheduled
Finance & Data

Data Analyst Agent

Answers plain-English questions over your databases and sheets, and sends leadership a weekly report.

DatabaseSheets
On demand
Legal & Compliance

Contract Review Agent

Reviews incoming contracts against your playbook, proposes redlines and points to the clause behind each change.

DriveDocs
Continuous
Legal & Compliance

Policy & Compliance Agent

Checks drafts and outgoing emails against your own policies, and alerts the right owner before anything goes out.

GmailDocs
Continuous
People

Recruiting & Onboarding Agent

Screens applications against each role's rubric, schedules interviews and builds a tailored onboarding plan for every hire.

GmailCalendarATS
On demand
Engineering

Engineering Docs Agent

Answers "how does this work?" from your code, RFCs and chat history, and flags docs that have drifted from the code.

GitDocsChat

Have a workflow in mind? We'll build it.

Any repetitive, knowledge-heavy work is a candidate. Tell us about it and we'll design the agent.

Start a custom build

How It's Built03

Every agent sits on the same private foundation: a knowledge graph of your company, built with GraphRAG and running on your servers. Add a new agent and it already knows your context.

  1. 01

    Connect your data

    Scoped, read-first connectors sync Workspace, CRM, databases and internal tools inside your cloud.

  2. 02

    Build the private graph

    Entity resolution turns documents, threads and meetings into typed entities and relationships.

  3. 03

    Design your agents

    Each agent gets its own job, tools and guardrails, plus the slice of the graph it needs.

  4. 04

    Deploy on your servers

    Everything runs in your VPC with your keys. Nothing is sent to external AI providers.

3a — What the graph stores

Agent answer · path followed

3b — Data flow, agent by agent

Source Process Store Output

Project Tags04

Each project in your company gets a tag. One prompt on a tag brings back the full project: what it is, who owns it, what has been decided, what is at risk and where every piece lives. You can share that brief with any team.

4a — One tag. The whole project.

Hand off a project in one prompt, not a week of meetings.

  • Tags are created automatically and stay up to date as work happens
  • The same tag produces a different brief for leadership, engineering or a new joiner
  • Briefs respect each person's access, so nobody sees what they couldn't already open
  • The result is documentation that writes itself and never goes stale
/
#atlas-q4-launchSynced 2 min ago · 31 sources

Owners

PRJMAK+3

Progress

68% · one dependency at risk

Key decisions

    14962

    Privacy05

    Data privacy is the reason this product exists. Your agents, your data and your models all run inside infrastructure you control.

    Nothing you connect is ever sent to an AI company. Not for answers, not for training, not to us.

    Typical AI tool

    Your company
    Your data
    Vendor's cloud
    Third-party model API
    Logs kept by the vendor

    Leaves your company at the first step, and copies stay outside it.

    OakCred Agents

    Your infrastructure
    Your data
    Agents in your VPC
    Private model
    Audit log you own

    Never leaves your infrastructure, and there is no connection back to OakCred.

    Runs on your servers

    We deploy to your AWS account, your cloud or your own hardware. Nothing is hosted by us.

    Never used for training

    Models run through Amazon Bedrock in your account, or as open-weight models on your GPUs.

    Access-aware

    Agents follow your existing permissions and SSO groups. If you can't open a file, the agent won't quote it.

    Your keys

    Data is encrypted in transit and at rest with keys in your own KMS. Rotate or revoke them at any time.

    Full audit trail

    Every retrieval, tool call and answer is logged in a store you own.

    Kill switch

    One switch pauses every agent and connector. Your infrastructure, your decision.

    Interface06

    Once the build is transferred, your teams use every agent from one clean workspace. You can chat, run workflows and review approvals. You operate it yourselves.

    agents.yourcompany.internal Private
    Ask any of your agents…⌘K

    Handover07

    We build it, deploy it on your servers and then transfer it to you completely: code, infrastructure and runbooks. We keep helping after handover for as long as you want us to.

    1. 01

      Discover

      We map your teams, workflows and the work that's worth automating.

    2. 02

      Connect

      We set up scoped, read-first connectors to your tools inside your cloud.

    3. 03

      Build the graph

      We resolve entities, tag projects and build hybrid graph and vector indexes.

    4. 04

      Build agents

      We design agents for your workflows and test them against real cases.

    5. 05

      Deploy

      We deploy on your servers with infrastructure as code. No dependency on us.

    6. 06

      Transfer

      You get the source code, infrastructure templates, docs and admin training.

    7a — After handover

    We stay close, without staying connected.

    • Help with new agents and workflows as your company grows
    • Model upgrades as better open models become available
    • Support for tuning, monitoring and incidents, on your terms
    • We only get access when you grant it, and you can revoke it whenever you like

    Stack08

    We use proven, inspectable infrastructure. The runtime isn't tied to any one model, so you can switch models without rebuilding your agents.

    05

    Interface

    Chat workspaceApprovalsAdmin console
    04

    Agents

    Amazon Bedrock AgentsTool use (MCP)Guardrails
    03

    Retrieval

    GraphRAGProperty graphVector indexEntity resolution
    02

    Models

    Bedrock in your AWS accountOpen-weight models on your GPUs
    01

    Connectors

    Google WorkspaceCRMDatabasesHelpdesk
    00

    Your infrastructure

    VPCKMS keysIAM and SSOInfrastructure as code

    Private by architecturenot by policy

    Custom by defaultbuilt around your workflows

    Groundedevery answer cites a source

    Ownedfull code transfer

    FAQ09

    Questions teams usually ask before a build.

    Does any of our data reach OakCred or an AI company?

    No. Connectors, the knowledge graph and every agent run in your environment. Models run through Amazon Bedrock in your own account or as open-weight models on your hardware. After handover there is no connection back to us.

    Which agents can you build?

    The catalogue is a starting point. We build agents for any repetitive, knowledge-heavy workflow, in operations, finance, sales, legal, people or engineering, and customise each one to your tools and processes.

    Which tools can the agents connect to?

    Google Workspace (Gmail, Drive, Docs, Sheets, Calendar, Meet and Chat), plus CRMs, helpdesks, ERPs, databases and internal tools, both as data sources and as tools agents can act on.

    Do we own the software?

    Yes. At handover you receive the full source code, infrastructure templates, documentation and admin training. You can run, change and extend it without us.

    What happens after handover?

    We keep providing assistance with new agents, model upgrades, tuning and incident help. We only get access when you grant it, and only for as long as you allow.

    Start a build

    Your agents. Your data. Your servers.

    Tell us which workflow you'd like to hand to an agent first. We'll come back with an architecture sketch for your stack.

    singhan@oakcred.com