How is Melaya Assistant different from a provider chatbot?

A provider chatbot primarily answers inside its own interface. Melaya Assistant works across the connectors, context, pipelines, and paired devices you authorize, while preserving model choice, human approval, and an auditable execution trail.

Can an enterprise use its own models and infrastructure?

Yes. Teams can use supported cloud providers, enterprise-hosted endpoints, CLI subscription runtimes, or local models through Ollama and LM Studio. Model choice is not tied to the workspace or the interface.

How does Melaya protect enterprise data?

Connector credentials, conversation history, and device sessions are tenant isolated. The assistant sees only explicitly selected connectors and context, sensitive actions can require human approval, and execution is recorded for review.

Can Melaya Assistant operate apps on a phone?

Yes, on a paired Android device. It can read and operate approved apps through Device Control even when those apps have no automation API, subject to the device allowlist and approval policy.

What are Safe, Payments only, and Autonomy modes?

Safe mode asks for approval on consequential writes. Payments only lets routine requested actions proceed while retaining payment approvals. Autonomy executes the user's requested automation within configured permissions and platform policy.

Can I see what the assistant used and what it cost?

Yes. Melaya records provider rounds, tool calls, outcomes, token usage, attributable cost, approvals, and execution spans so operators can inspect what happened instead of relying on a chat transcript alone.

02 / 03

One assistant.Every system. Under control.

Speak naturally to your business applications, your documents, your agent pipelines, and your Android phone. Choose the model, define the permissions, approve what matters, and inspect every action.

← Back to home

Melaya AssistantGoverned and ready

Safe modeYour model3 scoped systems

Which overdue invoices need attention, and what should we do next?

I found three overdue invoices in Odoo. I matched the accounts with recent email history and prepared the next actions for review.

Odoo · invoicesreadZoho Mail · historyreadFollow-up draftsapproval required

Ask across your workspace…

ScopedOnly what you select

MeasuredTokens and cost per round

TraceableTools, outcomes, approvals

PortableCloud, local, or your server

A chatbot knows what to say. Melaya knows what it may do.

Provider chatbots are useful destinations. Enterprises need an operating layer: defined access, verified actions, model choice, cost controls, and evidence after the conversation ends.

Enterprise requirement

Provider chatbot

Melaya Assistant

Business context

Provider chatbot

Files and text copied into one chat

Melaya Assistant

Static context plus selected live systems

Action

Provider chatbot

Advice or vendor-specific tools

Melaya Assistant

Scoped connectors, pipelines, and phone tools

Control

Provider chatbot

Prompt-level instructions

Melaya Assistant

Enforced modes and editable approval gates

Model strategy

Provider chatbot

One provider's catalog

Melaya Assistant

Cloud, enterprise, CLI, and local runtimes

Accountability

Provider chatbot

A conversation transcript

Melaya Assistant

Provider rounds, tool spans, outcomes, and cost

The four questions every enterprise assistant must answer.

Where does data go? What will it cost? What exactly happened? Can we change the deployment later? Melaya makes each answer visible and configurable.

Privacy by architecture

Conversation history, credentials, connector access, and device commands are tenant isolated. The assistant receives only the context and systems explicitly selected for that conversation.

Scoped access, encrypted credentials, isolated execution

Cost you can govern

Route each workload to the right model. Use cloud APIs, your existing CLI subscriptions, or local models. Provider-aware context limits prevent an oversized prompt from becoming an accidental bill.

Round-level tokens, attributable cost, bounded context

Observability, not trust

See provider calls, tool arguments, results, approval decisions, and final outcomes in the same trace. A claimed action can be checked against the execution evidence.

Replayable spans and truthful outcome records

Deployment without lock-in

Keep the interface and governance layer while changing the underlying model or runtime. Run in Melaya Cloud, on an enterprise server, through a supported CLI, or locally.

Model choice per conversation and workload

Ask the business. Not every application separately.

Select the systems needed for the task, then speak naturally across them. The assistant can correlate information, propose a next step, and route consequential writes through the selected control mode.

Explicit connector scope

A connected service is not automatically exposed. Users select the personal or project connectors available to each conversation.

Credentials stay server-side

Models receive tool schemas and results, never the raw connector secret used to execute the request.

From question to workflow

A successful request can become a repeatable pipeline with the same tools, permissions, and approval logic.

Melaya AssistantConversation scope

Only selected systems are available

Give it the brief. Keep the thread. Fit the model.

Static context anchors the conversation in your policies and objectives. Provider-aware memory preserves useful history without letting screenshots, tool results, or long threads overflow the model window.

STATIC CONTEXT

The facts that should not drift

Customer success policy.pdf

Escalation thresholds, approved language, account tiers, response targets…

PDF · TXT · MD

Conversation instructions
Add a stable brief for tone, policy, role, or task boundaries.
Document drops
Attach PDF, TXT, or Markdown references directly to the conversation.

MODEL-AWARE MEMORY

Long conversations without blind overflow

Context budget managed

Real context windows
Budgets adapt to the selected provider and model instead of assuming one arbitrary limit.
Whole tool rounds
Compaction preserves call and result pairing so provider payloads remain structurally valid.
Redacted summaries
Older context can be distilled asynchronously, with sensitive patterns redacted before model use and storage.

Choose how the request may execute.

The user requests the work. The selected mode defines where Melaya pauses. That policy travels with the turn into connectors, pipelines, and the paired phone.

MAXIMUM REVIEW

Safe

Pause consequential writes for a clear, editable human decision before execution.

Review before consequential action

ROUTINE FLOW

Payments only

Let requested routine actions proceed while keeping payment and financial commitments gated.

Payments always require approval

REQUESTED AUTOMATION

Autonomy

Execute the user's requested task within configured permissions, tool scope, device allowlists, and platform policy.

Permissions and policy still apply

Modes do not expand access. They change approval behavior inside the permissions already granted.

Compose a real request before you register.

Select connectors, choose a control mode, and type a task. This preview does not access or send data. Press Enter to continue into a free workspace.

New governed conversationReady

Model flexible

What should Melaya handle?

Ask across the selected systems or start with a proven request.

Speak to your AI. Let it work on your Android phone.

For founders, creators, operators, and mobile teams, Melaya Assistant can continue the requested task on a paired Android device, including apps that expose no automation API.

Real mobile interfaces

Open allowed apps, observe the visible screen, tap, type, swipe, and verify the next state.

Voice from anywhere

Use voice input and the mobile overlay to request work, follow progress, and respond to approvals.

Control remains on-device

The Android executor enforces the app allowlist, publishing gates, stop controls, and run watchdog where actions happen.

Explore Melaya Device Control

Melaya AssistantPhone online

Prepare replies to today's important comments and stop before publishing.

Working on your phone

Reading the approved app and preparing the next action

Speak to Melaya

Publishing approval stays on-device

A simple interface over a governed runtime.

The conversation stays natural. The execution underneath is explicit, bounded, and inspectable.

01 · ASK

State the outcome

Type, speak, or attach the context needed for the task.

02 · SCOPE

Select access

Choose the connectors, model, device, and operating mode.

03 · EXECUTE

Watch verified work

Follow tool calls, results, phone steps, and approval requests live.

04 · PROVE

Review the evidence

Inspect outcomes, cost, provider rounds, and the complete execution trace.

Before an assistant touches real work.

Clear answers for security, platform, operations, and individual users.

How is Melaya different from using ChatGPT, Claude, or another provider chat?

Melaya is the governed operating layer around the model. It connects the systems you explicitly select, can execute through tools and Android Device Control, preserves provider choice, enforces approvals, and records what happened. You can still use the models you prefer.

Can we run our own model or keep execution on our infrastructure?

Yes. Use supported cloud providers, enterprise-hosted endpoints, CLI subscription runtimes, Ollama, or LM Studio. Deployment and model strategy can evolve without rebuilding the user experience or governance layer.

Does connecting an application expose it to every conversation?

No. Connector availability is scoped. Users select the personal or project connectors for a conversation, and credentials remain server-side during tool execution.

What can we audit after a request?

Provider rounds, tool calls, arguments, results, approvals, token usage, attributable cost, and terminal outcomes are recorded as execution evidence, subject to your retention and deployment configuration.

Can the assistant control phone apps without an API?

Yes, on a paired Android device. Device Control operates approved visible interfaces through Android accessibility capabilities, with an on-device app allowlist and controls for sensitive actions.

Does Autonomy mode mean unrestricted autonomous access?

No. The user requests the automation, and Autonomy mode changes approval behavior for that request. Configured permissions, connector scope, device allowlists, safety controls, and platform policy continue to apply.

Turn conversation into governed execution.

Start free with the models and tools you already use. Add connectors, mobile execution, and enterprise deployment as the work becomes real.

Start with Melaya Assistant Read the documentation

Free to start · Human control built in · No model lock-in

Join the community