Introducing FRONTFACE

Turn what your company knows into a workforce.

FRONTFACE connects to your apps, learns what your company and each of your people know, and gives everyone knowledge agents that do the work. The more you use it, the more it knows.

A chat gives you an answer. A knowledge agent does the work.

Become a design partner See the benchmarks
Every app, data source and agent flowing into FRONTFACE's Knowledge Core Salesforce Gmail Calendar Slack Notion GitHub Coding agents Team agents Vendor agents Memory Company data Credits tool Knowledge Core
Task for Bob's agent · Sales rep

"Acme is unhappy. Sort it out."

  1. Read Acme's complaintGmail
  2. Checked their orders this quarterCompany data
  3. Found two open support casesSalesforce
  4. Issued a $300 credit, inside Bob's limitCredits tool
  5. Posted the update for the teamSlack

Done. Reply drafted for Bob's OK, every step logged.

Bob's credit steps are now a skill for his whole team.

Try a task

How FRONTFACE works

Learns. Works. Compounds.

Connect your apps once. FRONTFACE learns what your company and your people know, puts that knowledge to work, and gets better with every task.

Knowledge Core

Adventure Works is a sample company. The names and screens are examples.

It learns your company, from four sources.

What your people show it, what each app already offers, what agents discover, and verified open-source skills. All of it is curated before it reaches the Knowledge Core, so knowledge compounds instead of piling up.

  1. Your people show it

    Press record and do a task once, or correct a draft. It becomes your skill, and your team's once approved.

    Record buttonCorrectionsCoding agents
  2. Each app's own ways in

    The MCP servers, APIs and CLIs vendors already provide, or a custom CLI where none exists.

    MCPAPICLIVendor agents
  3. Agents explore

    A swarm of agents maps an app's screens and flows in a test account, and proposes skills.

    Test accounts only
  4. Verified open-source skills

    Proven skills in the open Agent Skills format, vetted before any agent may use them.

    Agent Skills formatVetted first

Curated

  • Tested on cases it hasn't seen
  • Approved by a person
  • Credited to whoever taught it
  • Re-checked when an app changes
  • Retired when it goes stale

Knowledge Core

  1. You"Acme is in a budget review" · only Bob
  2. TeamIssue a credit · from Bob's recording
  3. ProjectAcme fix ships in 4.3
  4. CompanyEscalation playbook · company skill

It puts that knowledge to work.

Every employee gets a knowledge agent, and teams get agents for their routine work. Together they are your knowledge workforce.

Your knowledge workforce · Sales Example
  • BBob's agentIssuing Acme's $300 creditWorkingChecked
  • JJennifer's agentDrafting Northwind's replyWorkingWaiting for Jennifer
  • SSales team agentSending this week's renewal check-insWorkingOwns this task
  • CCarol's agentPreparing Monday's review briefWorkingChecked

Ready to hand over nextPricing proposals

  1. Hand it a taskIn plain words, to your own agent or to your team's.
  2. It does the workIn your apps and with other agents, inside your permissions. It asks before it acts until its results earn your trust.
  3. See the whole workforceWhat every agent did, how well, and what to hand over next.

The more you use it, the more it knows.

Every task teaches it something. What one person teaches climbs to their team, their project and the whole company, still credited to them.

  1. You

    Bob corrects a draft: "Acme is in a budget review. Don't push a meeting."

    Kept for Bob
  2. Team

    Bob shares how he handles escalations, and Carol approves it.

    The sales team's agents use it
  3. Project

    The engineers' coding agents record that Acme's fix moved to 4.3, and why.

    Everyone on the Acme project knows
  4. Company

    The escalation playbook becomes a company skill, credited to Bob.

    Every new hire starts with it

Private notes never move. Only what a person agrees to share climbs a level.

The whole loop, end to end

Taught once. Used by everyone. Better every time.

Step 1 of 6 · Learns

Connect your apps once.

FRONTFACE plugs into the tools Adventure Works already uses, each through its own way in. Every agent sees only what its person may see.

  • GmailMCP
  • SalesforceAPI
  • Company dataMCP
  • CalendarAPI
FRONTFACE
  • SlackMCP
  • Credits toolCustom CLI
  • GitHubCLI
  • NotionAPI

Step 2 of 6 · Learns

It learns how the work gets done.

Bob shows it once how he handles a late order. The credits tool's own API and an exploring agent fill in what he didn't say.

Recording · Bob, Sales2:14

  1. Opens the late-order complaint
  2. Checks the order in company data
  3. Issues a $300 credit
  4. Tells the team in Slack
  • The credits tool's APICredits over $500 need a director
  • An exploring agentMaps the credit screens, in a test account

Draft skillHandle a late-order complaint

Step 3 of 6 · Learns

Nothing is used until it passes.

The draft is tested on past complaints, approved by a person and credited to Bob. His private notes stay his.

Curated

Handle a late-order complaintDraftApproved
  • Tested on past complaints it hadn't seen
  • Approved by Carol, Bob's team lead
  • Credited to Bob
  • Bob's private notes left out

Re-checked whenever the credits tool changes

Step 4 of 6 · The core

It lands in the Knowledge Core.

Filed at the level it belongs: Bob's team. Now any agent on the team can use it, inside its own person's permissions.

Knowledge Core · Adventure Works

  1. CompanyCredits over $500 need a director
  2. Project · AcmeAcme's fix ships in 4.3
  3. Team · SalesHandle a late-order complaint · from Bob
  4. You · Bob"Acme is in a budget review" · only Bob

Every agent on the sales team can use it now

Step 5 of 6 · Works

Any agent on the team can do it now.

A new complaint reaches Jennifer. Her knowledge agent handles it Bob's way, across her apps, and asks her before it acts.

Jennifer's agentUsing Bob's skill

Fabrikam: "Our order is late again."

  1. GmailReads the complaint
  2. Company dataFinds the order, six days late
  3. Credits toolPrepares a $300 credit, inside the limit
  4. SlackTells the team

Waiting for Jennifer's OKThe reply and the credit are ready

Step 6 of 6 · Compounds

And every task teaches it more.

Jennifer's correction sharpens the skill, and it climbs to the whole company, credited to both of them. Then the loop goes round again.

  1. Team skillHandle a late-order complaintFrom Bob's recording
  2. Team skill · v2+ Offer the credit before they askJennifer's correction, approved by Carol
  3. Company skillThe escalation playbookCredited to Bob and Jennifer. Every new hire starts with it.

Every task feeds what it learns next, so the next turn starts from more.

And round again

Adventure Works is a sample company. The names and screens are examples.

Use cases

Where teams start.

Five jobs teams hand to FRONTFACE first, from account managers to engineers. Each one plays through on its own, step by step.

Use case 1 of 5 · For account managers

Customer escalations

TodayA complaint touches five systems and waits for the one person who knows the account.

  1. GmailReads the complaint
  2. Company dataChecks this quarter's orders
  3. SalesforceFinds two open cases
  4. Credits toolIssues a $300 credit, inside the limit
  5. SlackUpdates the team

Done. The reply is drafted for approval, and every step is logged.

What it learnsHow your best rep handles escalations becomes the whole team's skill.

Benchmarks

Every agent sees only what its person may see. We measured it.

Agents act on what they can read, so access has to hold on every request. We test it in the open, with the method published.

340

Unauthorized results across 895 scored queries from five AI model vendors: the access rule written in the prompt, versus enforced on the query.

Read the analysis

0of 27,147

Executed queries returned a row the person may not see with enforcement on. Checked against an independent answer key, a negative control and nine seeded faults.

Read the analysis

17rule shapes

147 access roles, 17 rule shapes and 10 query structures. Testing the combinations, not just the query count, found three defects in our own rewriter.

Read the analysis

BlueKin's published analyses, October 2026, based on BIRD Mini-Dev and live PostgreSQL tests. All benchmarks

Demo

See FRONTFACE at work.

One month at a sample company: knowledge agents learning, working and compounding across ten apps, recorded from real runs on test accounts.

Watch the demo 5 minutes

Every company will have many agents.
The winners will own what those agents know.

FRONTFACE builds that knowledge from your apps and your people, and grows it with every task. We're opening it to a small group of design partners.

Become a design partner See the benchmarks