OGManager

Governed AI

AI agents that propose. A gateway that decides.

The model never holds credentials and never calls an external platform. It proposes typed tool calls; the action gateway checks tenant, permission, entitlement, credits, consent and risk, then asks a person or runs within your policy, verifies the result and logs every step.

Try the Command CenterHow we keep AI safe
app.ogmanager.com/app · Este Clinic · Gulshan

What do you want to accomplish?

Why did Gulshan bookings drop last week, and what should we do?

Here is what I am going to do

Plan · 4 steps
  1. Compare bookings week over week by sourceAnalytics agent
  2. Find the campaign that moved CPBAnalytics agent
  3. Recommend budget shift to the winning ad setStrategy agent
  4. Pause “Hydrafacial Eid offer”Campaign and Ads agent

Bookings · Gulshan · 7 days

142 −18%

Approval needed · Pause campaign

High risk

ROAS 0.7 over 5 days with ৳62,400 spent. Bookings from this ad set fell 41%.

ApproveRejectExpires in 23 h · Verified after execution

The action chain

read → analyze → recommend → confirm → execute → verify → log

Every write to an external platform, whether by a person, an agent or an automation, takes this path. Connectors expose execute() only to the gateway, so no other code path can change your ad account.

  1. 1Read

    Pull stored data and knowledge through scoped read tools. Never a live guess.

  2. 2Analyze

    Compare periods, find anomalies and the campaign or creative that moved the number.

  3. 3Recommend

    Propose a typed action with evidence, impact and the exact before and after values.

  4. 4Confirm

    Policy and risk decide: auto-run inside your limits, or an approval card for a person.

  5. 5Execute

    Only the action gateway can call a connector, with an idempotency key.

  6. 6Verify

    Read the external platform back to confirm the change actually landed.

  7. 7Log

    Append-only audit with who, why, model, approver, before, after and request ID.

Risk and approvals

You decide what the AI may do on its own

Each workflow and AI employee has an execution mode: recommend only, require approval, or auto-execute. Auto-execute must be switched on explicitly and covers only actions inside your policy.

  • “Maximum automatic budget change 10%”
  • “AI may spend up to ৳50,000 without approval”
  • “Budgets above ৳2,00,000 need the Marketing Director”
  • Approval re-checks that the target has not changed since the preview
Approvals inbox

Increase daily budget

Campaign and Ads agent · on behalf of Sudip

High

Before

৳30,000 / day

After

৳33,000 / day

  • Evidence: ROAS 3.4 over 14 days, CPB ৳410 (target ৳600)
  • Policy: changes above 10% need the Marketing Director
  • Target unchanged since preview · Meta Ads
ApproveRejectComment
Risk levels
RiskExamplesDefault handling
LowGenerate a report, caption or draft post; create an internal taskRuns automatically
MediumSchedule or publish a post, send a campaign message, update a CRM stageApproval; you can allow auto-run within limits
HighPause or enable a campaign, change a budget, create a campaign, refundApproval; MFA above your amount threshold
CriticalDelete a campaign, export customer data, 10× budget increase, payroll release, message all customersBlocked by default; Owner approval with MFA if enabled

13 specialist agents

A team of specialists, not one chatbot

An AI employee is an agent you configure: name, role, channels, permissions, working hours, confidence threshold and execution mode. Growth includes 4 AI employees, Pro 10 and Agency 25.

Agents and plan availability
AgentWhat it doesPlan
Analytics (AI Analyst)Answers questions with SQL and statistics, compares periods, finds anomalies and causesAll plans
ReportingBuilds dashboards and reports on request and writes executive summariesAll plans
RespondingInbox replies with intent, sentiment and lead score; verified prices and slots; booking; human handoffStarter+
StrategyFinds opportunities with impact, evidence, recommendation and actionGrowth+
Campaign and AdsBrief → audience → offer → content → budget → schedule → approval → launch; budget and pause proposalsGrowth+
AudienceSegments customers by purchases, revenue, frequency and engagement; syncs audiences to ad platformsGrowth+
Lifecycle and JourneyStages, segments, consent, sends, review requests, payment links and journeysGrowth+
ContentAds, posts, reels, email, SMS, captions and video scripts, grounded in your Business BrainGrowth+
SocialComment auto-reply, listening, trend radar, competitor tracking, best time to postGrowth+
OptimizationWatches KPIs, detects anomalies, recommends or executes within policyGrowth+
AutomationTurns a plain-language request into a workflow for your reviewGrowth+
CreativeBriefs, ideas, scores, storyboards, variants and creative QA for brand, price, claims and rightsPro+
Call CenterPost-call summaries, objections, outcome, QA score, coaching tips and tickets with SLAPro+

Business Brain

Grounded in your business, never in guesses

Every price, stock level, slot, balance and policy in a reply comes from a system record or a verified connector response. The model never invents a number.

  • Knowledge and memory

    Your profile, brand voice, goals and competitors, plus documents like price lists and treatment descriptions, searchable by role. Authorized users can view, edit and delete memory.

  • English, Bangla and Banglish

    The Command Center understands goals in English, Bangla and Banglish, and evaluation suites test Bangla intent in CI.

  • Injection-resistant retrieval

    Retrieved text is quoted as data, tools are allowlisted per agent and every output is schema-validated.

  • Human handoff

    Sensitive, low-confidence or policy-exception messages go to a person. After a human takes over, the AI stops replying in that conversation.

  • Replayable decision log

    Model, provider, prompt version, sources, tool inputs and outputs, policy result, approver, result and cost, with secrets redacted.

  • Your choice of provider

    OpenAI, Anthropic Claude or Google Gemini per organization, under terms that exclude training on your data.

  • One meter: AI credits

    Each job is estimated, reserved, run, then finalized at the actual amount, and the rest is released. Free includes 100 credits a month, Growth 5,000 and Agency 50,000.
    Learn more

  • Spend you control

    Owners are warned at 80% and 100%. Choose a hard stop or allowed overage, with a monthly spend cap per organization.

Bring your own AI client

The MCP server at mcp.ogmanager.com exposes the same governed tools to Claude, ChatGPT, Gemini and Copilot. Write tools like pause_campaign and update_budget create an approval request; nothing bypasses billing, permissions, consent or audit.

Read the MCP guide

Ask your business a question

Start a 14-day Growth trial with 5,000 AI credits and full governed actions. No card required.

Start free trialBook a demo