Decipher
Get Started
AI Agents & Workflow Automation

AI agents that finish the work. Not just talk about it.

Agents read the 500-page regulation. Scan 51 systems for PII. File the return. Triage the ticket. Draft the memo. They pick which tool to call next and only bring you what needs your call. Same tech behind Dcomply — India's DPDP-ready compliance platform.

51 auto-connectors in prod
Human-in-loop by default
Full audit trail on every step
Compliance agent · running Live
Scanned S3 bucket: 47,231 files
Found 12 unmasked Aadhaar numbers
Scanned Postgres: 3.2M rows
Awaiting DPO approval on 4 items
Filed GSTR-3B for Oct 2026
6 steps completed autonomously · 4 escalated to human
What Is An AI Agent

A worker, not a chat window.

An AI agent is a language model plus a set of tools it's allowed to use, plus a goal. You give it something like "close out this compliance filing" or "reconcile these two spreadsheets" and it plans the steps, picks the tools (an API, a database, a search, a file writer), executes them one by one, checks its own output, and only comes back to you when the job's done or when it needs a decision it can't make.

Chatbots have one loop: user talks, bot replies. Agents have many loops: think, act, observe, think again. Which is why they're slower and more expensive per query — but capable of finishing work a chatbot can't touch.

How an agent runs a task
G
Goal received
"Find all PII in our AWS S3 bucket and mask it."
P
Plan generated
List files → scan each → classify PII → mask → log → report.
T
Tools called in a loop
S3 API → OCR → PII classifier → mask → write log.
H
Human check on high-risk items
Ambiguous cases queued for DPO approval.
R
Report + audit trail
Every step logged. You can replay any decision.
Proof of Work

Dcomply: 51 agents running compliance

DPDP Act 2023 changed everything. Dcomply is our answer — compliance agents that read regulations, scan data, and file returns without a human in every loop.

The Problem

DPDP penalties up to ₹250 crore. Data scattered across 50+ systems.

Where is your PII? S3, GDrive, Postgres, Slack messages, WhatsApp exports, old CSVs on someone's laptop. Manual audits take weeks and miss things.

What We Built

51 auto-connectors + India-tuned PII discovery.

Each connector is an agent. Scans its source, classifies data, flags PII (Aadhaar, PAN, GSTIN, UPI, bank accounts), and reports. Another agent handles DSR (Data Subject Request) fulfilment end-to-end.

The Result

Discovery from weeks to hours. 80+ modules in one platform.

DPDP consent, DSR, PII discovery, breach notification, GSTR filings, ISO 27001 audit trail, sector-specific rules. The agents do the scanning, filing, and reporting. DPO signs off on the edge cases.

The stack that shipped it
Agent framework
LangGraph + Claude
Stateful multi-step flows
Tool calling
51 API connectors
S3, GDrive, Slack, Postgres…
PII classifier
Custom NER model
Aadhaar, PAN, GSTIN, UPI
Guardrails
Human-in-loop + audit
Every step logged, replayable
Six Agent Patterns We Ship

These are the ones that actually work in production.

We won't sell you an agent for something an agent isn't ready to do yet.

Compliance & Audit Agents

Read a regulation, cross-check your data, flag gaps, draft the filing, log the trail. Works for DPDP, GST, RBI KYC, SEBI, ISO 27001, sector rules.

Live in: Dcomply · 80+ modules
Document Processing Agents

Read invoices, contracts, IDs, medical reports. Extract structured data. Cross-check against your DB. Route by outcome. Handles handwritten, scanned, and multi-language.

Uses GPT-4o vision + custom OCR post-processing.
Ticket Triage & Routing

Read the ticket. Search past resolutions. Try to answer. If it can't, categorise + tag + route to the right queue with context. Cuts first-response time 60–80%.

Sits inside Freshdesk, Zendesk, Intercom.
Sales & Outbound Agents

Research the prospect. Pull context from CRM + LinkedIn + past emails. Draft a personalised outreach. Schedule follow-ups. Escalates to human on positive replies.

Feeds into your CRM. Never fully autonomous.
Data Reconciliation Agents

Match rows between two systems that don't quite agree. Bank statements vs. ledger. GSTR-2A vs. purchase register. Vendor list vs. payment records. Flag mismatches with reasoning.

Handles fuzzy names, dates, amounts.
Research & Reporting Agents

Pull data from 5–10 sources, cross-check, summarise into a report. Monthly compliance summaries, competitor tracking, market briefs. Same shape as Dcomply's monthly reports.

Delivered as PDF, Slack, or dashboard.
The Part Most Agencies Skip

How we stop agents from doing damage.

An agent that can call APIs can also delete rows, send bad emails, or file a wrong return. The four guardrails below are the ones we ship on day one — because we're the team that runs it after launch, and we don't want the on-call ping.

1
Whitelisted tool access
Agent can only call tools you explicitly allow. Everything else returns "not permitted".
2
Dry-run for destructive actions
Any delete, send, or file action shows a preview. Human confirms before it fires.
3
Hard spending caps
Model API cost cap. Tool call cap. Outbound message cap. Agent stops when it hits any of them.
4
Full replayable audit log
Every thought, tool call, input, output, decision. You can replay any step to see why it did what it did.
Process & Pricing

First agent live in 8–12 weeks.

Priced by phase. No surprises.

Week 1
Workflow audit

We map the task, the tools, the risk points, and the human-approval gates.

Free
Weeks 2–4
Prototype agent

One agent, one workflow, dry-run only. You watch it work on real data.

₹2–4 L
Weeks 5–12
Production build

Full integrations, guardrails, audit logging, human-in-loop, monitoring, staging + prod.

₹8–25 L
Ongoing
Operate

Model API costs, monitoring, tool updates, quarterly evals, incident response.

From ₹45k/mo
FAQ

What buyers ask us about agents.

An AI agent is a system that uses a language model to plan a sequence of steps, use tools like APIs or databases, and finish a task on its own. A chatbot answers questions. An agent does work. Read a 500-page regulation and pull out the applicable clauses. Scan 51 systems for personal data. Draft an email, check the calendar, book the meeting. The agent decides which tool to call next until the job is done.

RPA follows a script that a human wrote. It breaks the moment a screen layout changes or an input format shifts. An AI agent uses a language model to figure out what to do next, so it handles variation. Trade-off: RPA is faster and cheaper for rigid workflows. Agents are the right choice when the work involves reading unstructured data, making judgment calls, or working across many systems.

Prototype phase (one agent, one workflow) is ₹2–4 lakh. Full production build with multiple tools, guardrails, human-in-the-loop escalation, monitoring, and integrations is ₹8–25 lakh. Ongoing operations start at ₹45,000/month and include model API costs, monitoring, tool updates, and quarterly evals. Feasibility call is free.

A prototype agent runs on your data in 3–4 weeks. Production build takes 5–8 weeks after that. Total: 8–12 weeks. Complex multi-agent systems with 10+ tools take 12–16 weeks.

LangGraph for stateful multi-step workflows. LangChain agents for simpler patterns. AutoGen for multi-agent conversations. OpenAI Assistants API and Anthropic Claude tool-use for shipped features. We pick based on what the workflow needs — no religious preference.

Four layers. First, permissions — the agent can only call tools you explicitly whitelist. Second, dry-run mode — every destructive action is previewed and needs a human confirmation. Third, spending caps — hard limits on model API costs, database writes, and outbound messages. Fourth, full audit logs of every step the agent took, why, and what it changed.

Yes. Dcomply already has connectors to S3, GDrive, Slack, Postgres, MySQL, Zoho, HubSpot, Freshdesk, WhatsApp, Gmail, Notion, and 40+ more. If your system has an API, we can wire an agent to it. If it doesn't, we can build a browser-automation layer as a last resort.

Reading long documents and extracting structured facts. Cross-checking data across systems. Ticket triage and routing. Email drafting with context from your CRM. Compliance filings that follow rules. Research tasks that need multiple sources. Reporting that pulls from several places. What they're still bad at: tasks with high physical stakes or heavy real-time judgment. We won't sell you an agent for those.

Have a workflow you'd like to hand off?

Book a 30-minute call. Walk us through the work you'd like an agent to do. We'll tell you if it's a fit, what it costs, and what your team still needs to own.

Decipher Assistant
Typically replies instantly