Back to vendors
T

ToolJet

Also known as: ToolJet, tooljet.com, tooljet.ai, ToolJet Agent Builder, ToolJet Workflows, ToolJet MCP

Visit site
Entry price$199 per builder per month billed yearly ($249 monthly) for self hosted Team, the entry to the agent builder. A free community edition is available.Full pricing detail

Open source internal tools platform whose self hosted edition adds an agent builder: an AI agent node that picks its own tools inside visual workflows, woken by webhooks, schedules or app events, with SSO, RBAC and air gapped deployment.

ToolJet is a low code platform for internal tools, customer portals and back office applications, built around a visual app builder, a built in PostgreSQL database, a workflow engine and more than 100 data source connectors. Its agent layer sits inside the workflow canvas. An AI agent node takes a system prompt, a user prompt and an output schema, connects to a model the builder chooses, and can use any workflow node as a tool, such as a database query, a REST call to Slack, Gmail or GitHub, a JavaScript step or a ToolJet Database table. The agent decides which tools to call and in what order, within step, retry and timeout limits the builder sets.

Agents start the way workflows start. A ToolJet app can invoke one on a button click or form submission, an external system can call a typed webhook, including an asynchronous endpoint for long runs, and a scheduler fires on an interval or a cron expression in a chosen timezone. Workflows add branching, loops and child workflows around the agent, so an agent step can sit inside a longer process with deterministic steps on either side.

The agent builder ships only on self hosted ToolJet, which runs on Docker, Kubernetes, a customer VPC or on premises, with an air gapped option on the enterprise license. ToolJet Cloud carries the app builder and its AI assisted building features but not workflows or agents. Governance comes from the platform, with custom user groups, SSO through SAML, OIDC or LDAP, audit logs, and on enterprise plans SCIM, row level security and page, component and query level access rules.

ToolJet also offers an MCP server that lets a coding agent such as Claude Code, Codex or Cursor build and modify ToolJet apps, manage page and query permissions and administer users, authenticated with a personal access token and recorded in the platform audit log. On the agent side, there is no native approval step for agent tool calls, no memory layer across runs and no evaluation harness for agent behavior.

Vendor details

Canonical URL

https://tooljet.com

Category

Agent builder

Subcategory

Agent nodes inside an internal tools workflow engine

Funding status

Private and venture backed (ToolJet Solutions Inc, San Francisco). The pricing page reports 37,602 GitHub stars and 625+ contributors on the open source repository.

Company status

independent

Use cases & customers

Primary use cases

agent steps inside back office workflowsinternal tools and admin panels over existing databases and APIswebhook and schedule driven automation with an AI reasoning stepcustomer portals and business applicationsself hosted or air gapped internal app platforms

Target customers

enterprisemid-marketdevelopersIT teamsoperations teams

Deployment options

ToolJet Cloud on AWS (no agent builder)self hosted on Docker or Kubernetescustomer VPCon premisesair gapped (Enterprise add on)

Integrations

More than 100 data sources and SaaS connectors, including PostgreSQL, MySQL, MongoDB, REST and GraphQL APIs, Slack, Gmail, GitHub, Stripe, Weaviate and Hugging Face. Any workflow node can be handed to an agent as a tool. Model providers for the agent node include OpenAI, Anthropic, Gemini and Mistral. A REST API with personal access tokens, a GitSync API and an MCP server for coding agents extend the platform.

In practice

An operations team adds an agent node to a webhook triggered workflow so that inbound refund requests are checked against order data and routed to Stripe or a human queue.

An IT group runs the self hosted edition inside its own Kubernetes cluster so that agents, workflows and the apps that call them never leave its network.

A developer uses Claude Code with the ToolJet MCP server to generate an admin app against existing tables, then restricts its pages and queries to named user groups.

Agentic Index coverage score

10.5 / 14 capabilities · 75%

Integrations & Tool Calling Full

Any workflow node can be attached to the agent node as a tool, so agents act directly in outside systems. Tools include datasource queries against PostgreSQL, MySQL and MongoDB, REST calls to Slack, GitHub, Gmail and Twilio, JavaScript steps and ToolJet Database tables, over a catalog of 100+ connectors whose credentials stay server side. Examples include sending Slack messages, creating incidents and drafting or forwarding Gmail.

Sourcedocs.tooljet.com/docs/workflows/nodes/agentread 2026-10-02

Workflow Orchestration Full

A visual node based workflow engine runs multi step processes with logic nodes (JavaScript, If/Else, Loop), datasource and REST nodes, a workflow node for child workflow orchestration, and an agent node that plans its own tool calls within max steps, retries and timeout. Agents handle long running processes with retries and external waits, run in parallel across workflows.

Sourcetooljet.com/pricingread 2026-10-02

Knowledge Grounding & RAG Partial

The agent assembles context on each run, reading business data through the tools it is handed, such as database queries and API calls, and ToolJet describes its agents as context aware, fed by connected data. There is no index, embeddings layer or knowledge base built for agents to retrieve from. A Weaviate connector exists in the catalog, but the store is the customer's own and no retrieval feature is built on it.

Sourcetooljet.com/ai-agent-builderread 2026-10-02

Human Oversight & Guardrails Partial

Builders choose exactly which nodes an agent may call and cap it with max steps, max retries and a timeout, and workflow branching can route outcomes. ToolJet markets workflows that manage approvals and human in the loop steps, but its workflow nodes include no approval or human task node that pauses an agent action before it commits. ActionRail, a separate ToolJet open source project in public beta, holds agent tool calls for a human, but it is not part of the agent builder.

Sourcedocs.tooljet.com/docs/workflows/nodes/agentread 2026-10-02

Security, Identity & Governance Full

Access runs through RBAC with predefined and custom user groups and SSO through Google, GitHub, OIDC, LDAP and SAML, and Enterprise adds SCIM provisioning, group sync, row level security and page, component and query level access rules. ToolJet undergoes regular SOC 2 Type II audits and keeps a trust center at trust.tooljet.com. For ISO 27001, ToolJet says it follows the standard rather than holding the certification. Audit logs are kept 14 days on Team and unlimited on Enterprise.

Sourcedocs.tooljet.com/docs/security/complianceread 2026-10-02

Observability & Auditability Full

Workflow run logs with node level visibility show each tool node the agent invoked and the parameters it passed. Agent runs can be debugged by clicking to inspect them, platform audit logs record acting user, action and timestamp, and self hosted instances export OpenTelemetry. There are no reasoning traces from inside the agent node.

Sourcetooljet.com/pricingread 2026-10-02

Memory & State Persistence Partial

State persists inside a run, where agents retain context and continue reasoning across extended processes with waits and retries, and only the agent node's final result passes to later nodes. No memory lasts across runs. The agent can query ToolJet Database tables, which hold the application's data.

Sourcedocs.tooljet.com/docs/workflows/nodes/agentread 2026-10-02

Deployment & Data Residency Full

Self hosted ToolJet runs on Docker, Kubernetes, the customer's VPC or on premises, with multi instance and air gapped deployment on the Enterprise license, and the agent builder is available only on self hosted. ToolJet Cloud runs on AWS EKS.

Sourcetooljet.com/pricingread 2026-10-02

Prebuilt Agents, Templates & Packs Partial

App templates for lead management, KPI dashboards, inventory, leave management and applicant tracking can be adopted, but none is an agent or an agentic workflow. Scheduling, Slack task, purchase order, campaign and refund agents appear as worked examples the customer assembles, not as installable prebuilt agents.

Sourcetooljet.com/ai-agent-builderread 2026-10-02

Triggers & Channel Coverage Full

A webhook trigger with typed parameters lets external systems invoke a workflow and its agent, an asynchronous webhook endpoint serves long runs, and a scheduler fires on an interval or a cron expression in a chosen timezone on paid self hosted tiers. App events such as button clicks, form submissions or data changes also start agents.

Sourcetooljet.com/ai-agent-builderread 2026-10-02

Model Flexibility & Routing Full

Each agent node is linked to an AI model datasource the builder configures, with OpenAI, Anthropic, Gemini and Mistral supported and 10+ vendors claimed, so different models can serve different use cases. Temperature, max tokens and top P are set per node. Enterprise adds custom AI model options, an on premises AI gateway and bring your own keys for ToolJet AI.

Sourcedocs.tooljet.com/docs/workflows/nodes/agentread 2026-10-02

APIs, SDKs & MCP Extensibility Full

ToolJet has an API for the platform itself, with personal access tokens, user management APIs and a GitSync API, and webhook endpoints invoke workflows. An MCP server, authenticated by a workspace scoped personal access token, has a published list of supported tools. Those tools create and modify apps, set page and query permissions and manage users.

Sourcedocs.tooljet.com/docs/build-with-ai/mcp/overviewread 2026-10-02

Testing, Debugging & Optimization Partial

Agent runs can be debugged visually by clicking to inspect them, and plans include run logs and AI auto fix for apps. There is no evaluation harness, scored test case or quality gate for agent behavior. Automation testing covers the development lifecycle of apps.

Sourcetooljet.com/ai-agent-builderread 2026-10-02

Browser & Computer Use Not documented

The agent has no browser, desktop or remote computer control. Agent tools are datasource queries, REST calls, JavaScript and ToolJet Database. A customer's own coding agent with browser access can open an app it built through MCP, but that agent belongs to the customer, not ToolJet.

Sourcedocs.tooljet.com/docs/workflows/nodes/agentread 2026-10-02

The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded

Pricing

$199 per builder per month billed yearly ($249 monthly) for self hosted Team, the entry to the agent builder. A free community edition is available.

Builder seats, with end user caps per tier and a monthly AI credit allowance per builder. Add on AI credits are sold separately.

Free tierTrial available

Included quota

Self hosted Free: 2 builders, 50 end users, 10 apps, 2 agents, 100 AI credits a month. Self hosted Team: per builder seats, 100 end users, unlimited apps and agents, 3,000 AI credits per builder a month, 14 day audit logs.

What is public

Both the Cloud and self hosted plan tables, per builder prices, end user and app limits, AI credit allowances, the add on credit price and full feature comparisons are published. Enterprise is custom.

Billing mechanics

Per builder subscription, monthly or yearly with a 20 percent yearly saving, plus a shared workspace AI credit pool. Self hosted Team is bought as a license through a checkout on the pricing page, and Enterprise is quoted.

Cost watchouts

The agent builder is self hosted only, so the customer also carries infrastructure and operations cost. End user caps (50 on Free, 100 on Team) push larger internal audiences to Enterprise for unlimited end users. SCIM, row level security, granular access rules, BYOK and air gapped deployment are Enterprise only. Model calls made by the agent node run on the customer's own model provider account.

Variable cost rationale

ToolJet bills builder seats at fixed published prices, so the bill grows with the number of builders and with the tier that end user caps require. AI credits are the metered part. Each builder gets a monthly allowance that resets without rollover, and extra credits cost $1 per 100. Self hosting, which the agent builder requires, moves infrastructure cost to the customer, and model calls made by agents run on the customer's own model provider bill.

Additional watchouts

The Cloud prices are app builder prices only. Agents need the self hosted edition.

Overage / add-ons

AI credits reset monthly and do not roll over. Add on credits cost $1 per 100 and last a year, and exceeding end user caps requires a higher tier.

Sales call required

Mixed (some tiers require a call)

Free / trial

Free self hosted community edition with 2 agents and 50 end users, and a free Cloud plan without agents. A 14 day trial is offered.

Lowest paid plan

Cloud Basic at $23 per builder per month, billed yearly, without agents. The agent builder starts at self hosted Team, $199 per builder per month billed yearly ($249 monthly).

Commercial notes

The Team license is bought through a self serve checkout, and Enterprise through a sales call. A free community edition supports open source self hosting.

Key ambiguities

The cheapest paid plan (Cloud Basic at $23) does not include workflows or agents, so the paid entry price for the agent builder is the self hosted Team license. Whether agent node model calls ever draw on ToolJet AI credits, rather than the customer's own model datasource, is not stated.

Missing data

Enterprise floor, and whether agent runs consume AI credits.

Agentic Index verified 2026-10-02

Alternatives to ToolJet

The closest documented capability profiles to ToolJet among agent builders tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.

  • Activepieces11.5 / 14Fuller documented coverage on Knowledge Grounding & RAG and Human Oversight & Guardrails
  • Budibase11.5 / 14Fuller documented coverage on Knowledge Grounding & RAG and Human Oversight & Guardrails
  • Celigo10.5 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs
  • Airtable11.0 / 14Fuller documented coverage on Knowledge Grounding & RAG and Prebuilt Agents, Templates & Packs
  • Bika.ai10.0 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs
  • Cohere North11.0 / 14Fuller documented coverage on Knowledge Grounding & RAG and Human Oversight & Guardrails

Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded

Contact us

Found a vendor we missed? Have feedback on the index? We'd love to hear from you.