Canvas Labs

What is Canvas Labs?

Canvas Labs is an AI application laboratory operated by Isometry AI Private Limited in Bangalore, India. The team starts from business challenges, data, and costs, and approaches the design, development, deployment, and handover of AI systems as a complete project.

It is not a single chat tool; its current products include Canvas Connectors and Sketch, while the Fine-Tuning Lab was discontinued in earlier stages. The lab also develops Context Graph and industry-specific AI systems in collaboration with companies.

Product portfolio and current status

Main functions of Canvas Connectors

Built-in proxy tool

• File tools: Read, create, overwrite, and precisely edit files to which the agent has access.
• Execution tool: Run commands within a sandbox, thereby limiting interactions with the production system.
• Search tools: search for file contents, find files by name, and retrieve live web pages.
• Web tool: Reads text from web pages and handles websites that require script rendering.

Unified connection to over 3,000 applications

Connectors use a consistent approach to authorization, parameter structure, and auditing in order to link messages, documents, projects, customer relationships, customer service functions, data warehouses, and code platforms. Examples of such connectors include Slack, WhatsApp, Teams, Gmail, Notion, Jira, Salesforce, HubSpot, ServiceNow, Snowflake, and GitHub.

Permissions, approvals, and auditing

Administrators can set permissions based on organization, agent, user, and action; for example, they can restrict access to a customer relationship system to read-only mode or allow messages to be sent only to specific Slack channels. Actions that have side effects may require manual approval, and agents cannot expand their authorities on their own.

Each call is logged, including the tool used, parameters, authorized user, time, and response; rejected calls are also recorded. The audit data can be queried and exported to the enterprise logging system.

MCP and API integration

Connectors can be used to connect MCP servers to Claude Code, Cursor, or other MCP clients; their standard APIs can also be invoked by any proxy framework. They serve as a tool and governance layer, and there is no requirement for users to change their proxy framework.

Usage process of Canvas Connectors

1. First, test the built-in tools and a few connectors in a free sandbox.
2. Define the read-only, write, message-sending, and command-execution boundaries for specific agents.
3. Connect to a low-risk application and save the credentials in a controlled key system.
4. Execute a rollback action via MCP or API, and check the audit logs.
5. Add manual approval for write, send, delete, and create commands.
6. Verify reliability, invocation costs, error retries, and third-party throttling through pilot testing.
7. Choose between managed, self-hosted cloud, self-managed, or fully isolated production deployment.

Connector deployment and data boundaries

The platform indicates that customer data remains in the original business system; the Connectors tool processes the data and returns the results to the proxy. In the self-hosted and fully isolated modes, credentials and audit logs can be kept within the customer’s network.

What can Sketch do?

• Answer organizational questions and carry out cross-system tasks via Slack, WhatsApp, and email.
• Use the Context Graph to link people, projects, historical messages, and business decisions.
• It connects over 3,000 integrations and enables the execution of an unlimited number of workflows and scheduled tasks.
• Teams are allowed to use their own keys from Anthropic, OpenAI, or other model providers.
• Credits are consumed based on the complexity of the task; reusing contexts can reduce this consumption through caching.

Sketch price

The hosted version of Sketch is charged based on credits, with no additional fee per user; the official website currently offers only monthly billing. Any unused credits are reset at the end of each month and are not carried over to the next period.

How to consume credits

• Examples of quick tasks range from 10 to 200 credits, such as generating summaries, updating fields, or sending follow-up emails.
• Examples of complex workflows cost between 100 and 300 credits, such as compiling reports across multiple systems.
• A complete project example costs between 500 and 750 credits; for instance, auditing the sales pipeline and creating dashboards.
• The actual consumption varies depending on the model, context, tool calls, and task complexity.

Sketch trial, cancellation, and refund

Sketch offers a 30-day trial period without the need for a credit card; if no upgrade is made, the account becomes inactive automatically. The subscription can be canceled at any time by contacting support via the account email, and it remains valid until the end of the current billing cycle.

• A refund can be requested within 7 days from the date of the first deduction; refunds can also be considered in cases of repeated deductions or system errors.
• After more than 7 days, refunds are generally not provided for partially used billing cycles.
• After the account becomes invalid or is canceled, the workspace data is retained for 90 days before being permanently deleted.
• The annual subscription has not yet been launched; the page only indicates that discounts are planned to be offered in the future.

Prices for Canvas Labs and Connectors

There is no fixed public price for custom projects at Canvas Labs; the scope, timeline, SLAs, delivery procedures, as well as rules regarding cancellation and refunds are specified in the individual order. Canvas Connectors offers a free sandbox environment, while production use is billed based on the number of calls made, in addition to a fee for the deployment platform.

API, GitHub, and open-source status

Privacy, security, and copyright

• The general policies apply to the Canvas Labs website and to our overall relationship; if products such as Sketch have their own separate policies, those product-specific policies shall take precedence.
• The platform may handle names, email addresses, companies, roles, account identifiers, customer-shared documents, messages, error logs, and account data.
• Payments are processed by Stripe and Razorpay; Canvas Labs does not store complete card or bank account details.
• The general policy states that Sketch’s real-time messages are not retained after processing, whereas project-related materials are stored, deleted, or returned in accordance with the terms of the contract.
• Once an account is deleted or a request is made, personal data is usually removed within 30 days, except for those data whose retention is required by law.
• Public measures include TLS, static encryption, access control, and regular security audits, but they do not guarantee zero risk.
• The customer retains ownership of the content entered and grants a limited license for processing it as necessary to provide the service.
• AI outputs may contain errors, biases, or omissions; significant business, medical, and compliance decisions must be verified manually.

Summary

Canvas Labs is suitable for companies that wish to integrate AI agents into their actual business systems, within an organizational context and within processes that allow for auditing. Canvas Connectors handle tool invocation and governance, while Sketch facilitates collaboration across different channels. Before making a purchase, it is necessary to check the number of invocations, the cost of the deployment platform, the amount of Sketch credits available, as well as the level of privacy protection offered by the product, in addition to reviewing the open-source terms and the availability of the code in the relevant repositories.