The platform

AI agents connected to everything your company already has.

IAToolkit puts agents to work on each company's documents, data and systems. You design them without code, use them in chat or from Claude, automate them, and run them with control over access and cost.

24
modules
100+
capabilities
6
model providers
3
ways to use it
Architecture

The platform at a glance

People and systems arrive through different channels. IAToolkit puts agents to work on the company's knowledge, reasons with the right model and acts on internal systems.

Capabilities

What each module can do

Twenty-four modules grouped in five layers.

Conversation

What users see and use every day: the chat, the assistants and the company memory.

Chat

An enterprise chat where each person talks to their company's agents about their own data.

  • Pick agent and modelUsers talk in natural language to the assistant they need, with the model they prefer.
  • Visible reasoningShows how the model is thinking while it builds the answer.
  • MultimodalAccepts attached documents, images, audio and video as part of the question.
  • Answers you can useDelivers tables, structured output and Excel files ready to download.
  • Long conversationsCompacts the history in the background to keep context without losing quality.

Prompts

Design specialized assistants without code.

  • Visual editorBuild prompts with custom fields the user fills in when starting a conversation.
  • PlaygroundTest each assistant before publishing it to users.
  • Fine control per assistantEach assistant sets its model, reasoning level, allowed tools and how much context it receives.
  • Versions and metricsKeeps versions and shows usage and token consumption per prompt.
  • Structured input and outputTakes attachments and returns data in the format another system expects.

Memory

The company memory: what the company knows, compiled and always up to date.

  • Feeds itselfCompiles knowledge pages from Google Drive, meetings and company documents.
  • Living memoryUpdates when its sources change, without blocking users.
  • Asks for what is missingDetects information gaps and asks the people who know.
  • Audience-based accessEach page reaches only the people who should see it.
  • Memory consoleDocuments, wiki and memory settings in one place.
  • Available everywhereAgents use it in chat, and clients like Claude use it over MCP.

Knowledge and data

The sources behind the answers: documents, wikis, knowledge tables and databases.

RAG and documents

Semantic search across all company documents.

  • Understands any documentParses PDF, Office files and images, including visual content, with Docling.
  • CollectionsOrganizes documents by topic and searches several collections at once, with ranking.
  • Automatic ingestionPulls documents from Google Drive and cloud storage with no manual upload.
  • Configurable embeddingsEach company picks its embedding model.
  • RAG panelManage documents, collections and sources from the console.

Wiki

Knowledge wikis that agents read and cite.

  • Wikis per companySearchable knowledge pages maintained by the team.
  • Read by agentsAgents search and read the wiki to answer with official information.
  • Per-wiki permissionsControl who can see each wiki and each page.
  • Exposed over MCPClaude and other MCP clients can query them directly.

Knowledge Tables

Knowledge tables: structured data that agents query precisely.

  • From a sample fileCreate a table from an Excel or CSV file and the platform infers its structure.
  • Fed by connectorsConnectors and pipelines keep it up to date, with load health and history.
  • Queryable with SQLAgents answer exact questions about the table.
  • Shared over MCPPart of the SQL catalog available to external clients.

SQL and Data Explorer

Agents that query company databases in natural language.

  • PostgreSQL and MySQLConnects to company databases, including on-premise ones through IAToolkit Bridge.
  • Schemas with meaningEditable, described schemas so the model understands the data.
  • On-demand schemaThe agent loads only the tables it needs, even in very large databases.
  • Safe queriesRead-only access and reinforced checks before running.
Learn more

Agents and execution

How agents work: models, tools, MCP, background work, automation and verification.

Agents

Configurable agents that work in chat, in the background or from other tools.

  • Their own configurationEach agent has its own model, tools, context policy and time limit.
  • Run wherever neededIn chat, in the playground, over MCP, as a scheduled job or inside a loop.
  • System agentsThe platform ships ready-made agents each company can turn on.
  • Decision agentsReturn a structured decision another system can consume.
  • SkillsReusable capabilities added to any agent.
  • Group accessEach agent is available only to authorized groups.

Tools

The agents' hands: tools to query and act on systems.

  • Native, HTTP and packsReady-made tools, calls to any API and themed packs.
  • Bridge to internal APIsReaches company systems that are not exposed to the internet.
  • Smart routingEach agent receives only the tools relevant to the question.
  • Policy and permissionsDefine which tool each agent and each user can use.

LLM models

Multi-model: pick the best model for each task, without lock-in to one provider.

  • Six providersOpenAI, Gemini, Anthropic, DeepSeek, OpenRouter and compatible open source models.
  • Model catalogPrices, capabilities and context window for each model, always current.
  • Keys per companyEach company can use its own credentials.
  • Reasoning and multimodalityUses each model's reasoning and its image, audio and video input.

MCP

IAToolkit as an MCP server: Claude and other clients work with the company's resources.

  • A single URLConnect with OAuth to use tools, agents, wikis, memory and data from Claude.
  • MCP AppsInteractive interfaces that open inside the MCP client.
  • AuditLogs and telemetry for every call, by user and tool.
  • MCP clientIAToolkit agents also consume external MCP servers.
  • Loops as toolsA verified loop can be invoked from any MCP client.
Learn more

Async tasks

Reliable asynchronous execution for heavy work.

  • Supervised workersQueues and workers that process in parallel and recover on their own.
  • Everything in the backgroundDocument ingestion, pipelines, jobs and agents without blocking the user.
  • API invocationsOther systems launch agents and receive the result.
  • Task monitoringStatus, retries and reconciliation for every task.
Learn more

Scheduled jobs

Agents and processes that run on their own, on a schedule.

  • Cron-style schedulingSchedule agents, native jobs and loops.
  • Platform schedulerJobs that run for every company from one place.
  • Run historyResults, failures and metrics for every run.
  • Native jobsReady-made processes shipped with the platform, such as data synchronization.

Pipelines

Pipelines that turn data and documents into structured information.

  • Cloud sourcesTake files from storage, Drive and other sources.
  • Configurable stepsParsing, LLM extraction, tools and structured schemas.
  • SQL output and write-backWrite results to tables or back to the source systems.
  • At scaleProcess thousands of documents with workers, with monitoring and reconciliation.

Loops

Loops: the agent does the work, checks it and fixes it until it meets the bar.

  • Run, verify, repairThe agent produces a result, rules verify it and the loop repairs it.
  • Graph stepsMulti-step flows with an LLM reflector and planner.
  • VerifiersRules on the result, SQL queries and cross-list checks.
  • Controlled budgetCost and attempt limits per run.
  • Human approvalReview before the result is published.
  • IntegratedScheduled as jobs and invoked over MCP.

Evaluations

Evaluations to change models and prompts based on evidence.

  • Real casesStores the full inputs of real runs so they can be replayed.
  • ComparisonMeasures how an agent behaves before and after a change.
  • Evaluations appCase sets and results in the console.

Connectors and communications

Connectors that link IAToolkit with the company's systems.

  • Google Drive and GmailDocuments and email with OAuth shared across the company.
  • IAToolkit BridgeSecure access to on-premise databases and APIs.
  • CommunicationsMessages over WhatsApp and email, including campaigns.
Learn more

Platform

What makes IAToolkit multi-company, secure and operable.

Multi-company configuration

One platform, many companies: each one configured to fit.

  • Company as configurationEach company's branding, models, tools, prompts and sources defined in YAML.
  • Editor with validationEdit the configuration from the console and validate it before applying.
  • Export and importMove configuration between environments and companies.
  • BlueprintsTemplates to create a new company ready to use.
Learn more

Users and access

Secure access at company, group and resource level.

  • Flexible loginLocal account, corporate SSO or Google.
  • Invitations and sign-upSign-up policies with bot detection.
  • API keysSystem-to-system integrations.
  • Groups and permissionsGroup access to agents, tools, wikis and data.
Learn more

Monitoring and analytics

Full visibility into how AI is used across the company.

  • Usage dashboardUsers, queries, consumption and cost per company.
  • Agent and tool monitoringPerformance and errors of agents, tools, MCP and tasks.
  • InspectorReview each conversation with its attachments and tool calls.
  • Usage analyticsAdoption and consumption metrics for operations.
Learn more

Internationalization

Ready to run in several languages.

  • Translated interfaceChat, administration and public site in several languages.
  • Language selectorEach user picks their language.
  • Local formatsDates and numbers follow the user's browser.

Hosted

The platform as a service: set up in minutes, usage-based billing and operations handled by IAToolkit.

Self-service onboarding

A new company signs up on its own and is up and running.

  • Self-service sign-upRegistration, plan choice and payment in one flow.
  • Ready to useThe company starts from a blueprint, with a getting-started guide.
Learn more

Billing and payments

Usage-based billing, transparent for the customer and the operator.

  • Rate cards per modelPer-model prices, updated from OpenRouter.
  • Closings and invoicesDaily and monthly closings with automatic invoicing.
  • Card paymentCustomers pay online.
  • Flexible plansSeats, support and credits.
Learn more

Hosting Control Center

Hosting Control Center: the console to operate the whole platform.

  • Companies and installationsCompany sign-up, contracts and licenses.
  • Usage and billingConsumption, invoices and coverage for each customer.
  • Central operationsModels, scheduled jobs and system agents for the whole platform.
  • InfrastructureWorker status and operational controls.
Learn more
Editions

Three ways to use it

The same platform, from the open source core to the managed service.

Open source, MIT license

Community

The core: chat, assistants, RAG, SQL, models, tools and the multi-company base. To validate your first case with your technical team.

Explore Community
On your infrastructure

Self-hosted

The full platform on your servers: MCP, automation, connectors, memory, monitoring and admin consoles.

See Self-hosted
Managed platform

Hosted

IAToolkit as a service: your company signs up, picks a plan and is up and running. We operate the platform.

See pricing
Evolution

A product that moves forward every week

Since January 2025 IAToolkit has added capabilities continuously, with releases shipped almost daily.

  1. Jan – Sep 2025

    Open source core

    Chat, assistants, RAG, SQL queries, OpenAI and Gemini support, tools and multi-company configuration.

  2. Sep – Oct 2025

    Consolidation

    Company registry, per-company branding, a redesigned chat and sessions with access logs.

  3. Nov 2025 – Feb 2026

    Knowledge and Enterprise

    Collections and ingestion sources, workers for background work, the admin console, several languages and the Hosted version.

  4. Feb – Apr 2026

    Automation

    Data pipelines, HTTP tools and packs, Anthropic, OpenRouter and open source models, and the Google Drive connection.

  5. May – Sep 2026

    Open ecosystem

    IAToolkit as an MCP server with audit and MCP Apps, knowledge wikis, WhatsApp communications and usage-based billing.

  6. Sep – Nov 2026

    Operational intelligence

    Knowledge Tables, evaluations, living memory, system and decision agents, group access, and Loops that verify the agents' work.