Our product

AI Consulting · Manas.Ai

An AI assistant that
knows your codebase.

Answers, code reviews, new code and root causes, grounded in your own code, documents and data, and running on your own infrastructure.

6

capabilities in one assistant

5

guardrail checks on every request

0

third-party AI services involved

The problem

Large, long-lived codebases
slow every team down.

01

Knowledge sits in a few heads

When the people who built it are busy or have left, every question waits for them.

02

Onboarding takes months

New developers read code for weeks before they can change anything safely.

03

Every change carries hidden risk

A controller, a view and a query depend on each other in ways nobody wrote down.

04

Business rules live everywhere

The code, the BRDs, the RFCs and the database each hold part of the answer.

What it does

Six capabilities
in one assistant.

01

Knowledge Chat

Ask how anything works. Each question is routed to the right sources, and answers draw on code and business documents and cite where they came from.

Cited answersCode + documentsDiagrams
02

Code Review

Reviews a change against your team's own guidelines before a person has to, with each issue tied to a line and a suggested fix.

Your guidelinesLine-level issuesSecurity findings
03

Code Generation

Writes PHP, JavaScript and SQL that use your real tables, columns and functions, then checks the result against them and reviews it.

Real schemaHouse styleSelf-reviewed
04

Ask Your Data

Plain-English questions become a single read-only SQL query against your database, with the result summarised for a business reader.

Read-onlyOne SELECTSummaries
05

Log Analyzer

Upload a log file and get its errors grouped and de-duplicated, each with a likely root cause traced to the code behind it.

Grouped errorsRoot causeCode context
06

Documents

Upload RFCs, BRDs and specifications, then ask about them alongside the code they describe.

RFCsBRDsSpecifications

How a request flows

Every capability passes
through the same guardrails.

In

Request

A question, code, a log file or a document.

Guardrail

Input checks

Prompt injection, blocked topics and requests for credentials are stopped. Personal data is masked.

Capability

Grounded in your sources

  • Knowledge Chat
  • Code Review
  • Code Generation
  • Ask Your Data
  • Log Analyzer
  • Documents

Guardrail

Output checks

Secrets and account numbers are redacted. The answer's quality is checked and its sources attached.

Out

Answer

With its sources, logged and ready to share.

Guardrail matrix

Admins switch each check on or off for each capability.

Access matrix

Admins create every account and decide who can use which capability.

Inference log

Every run is recorded: who asked, tokens, timings, the guardrail outcome and the user's feedback.

Made for teams whose code can't leave the building.

Manas.Ai deploys on your servers and talks to a model server you run. Your code, documents and data stay inside your network.

It is built by an InfoSec-led team, for banks and other regulated organisations, where every answer has to be traceable.

Built for control

Security and audit
designed in, not added on.

Runs on your infrastructure

Deploys on your servers and uses a model endpoint you choose from an approved list.

Guardrails both ways

Unsafe prompts are blocked on the way in, and sensitive values are redacted on the way out.

Access per person

There is no self-registration. Admins create accounts and grant capabilities one by one.

Every answer traceable

Each answer keeps a record of what was retrieved and how it was built.

Sharing with an audit trail

Share a conversation with a teammate. Every view of it is recorded.

Read-only on your data

Ask Your Data never writes, and it does not keep the rows it returns.

Under the hood

One stack,
all of it on your servers.

Interface

ReactViteTailwind CSSMermaid diagrams

Application

Python · FastAPISQLAlchemy · AlembicSQLite for users, settings and logsGuardrail pipeline

Retrieval

ChromaDB vectorsBGE-M3 embeddingsBM25 keyword searchCode graphRelevance check

Model

Your model server: Ollama or MLXOpenAI-compatible API

Your sources

PHP · JavaScript · BladeBRDs and RFCsOracle, read-onlyLog files

Who benefits

What changes for developers
and team leads.

Developers

  • Answers without waitingAsk how a flow works instead of booking time with the one person who knows.
  • Productive soonerNew joiners explore the codebase with answers that cite the code.
  • Code that fitsGenerated code uses your real tables, functions and conventions.
  • Faster root causeLog errors are grouped and traced back to the code behind them.

Team leads

  • Consistent reviewsEvery change is checked against your guidelines before a person reviews it.
  • Less key-person riskWhat lives in a few heads becomes answerable for the whole team.
  • VisibilityUsage, feedback and guardrail outcomes on one dashboard.
  • Tuned to your teamAccess, guardrails and every prompt are set from Admin, without a redeploy.

Getting started

Start with one codebase.

01

Scope

Pick one repository, its business documents and, if you want Ask Your Data, a read-only database account.

02

Deploy

We install Manas.Ai on your infrastructure and connect it to your model server.

03

Index

The code, templates and documents are indexed so answers can cite them.

04

Use and tune

Your team starts asking. Admins adjust access, guardrails and prompts as you learn what works.

See it on your code

Want to see Manas.Ai answer questions about your own codebase?

Book a demo