Budibase

Agents take authenticated action in outside systems: any table, REST API request, integrated app, knowledge source or automation in the workspace can be granted as a tool, Budibase discovers operations from connected data sources, and the custom REST connector carries configurable auth. The worked example lists, gets and updates ticket rows. Budibase agent building 101, agent tools and permissions docs and pricing table, 2 Oct 2026.

Sourcedocs.budibase.com/docs/agent-building-101read 2026-10-02

Multi step and multi agent execution: the automation builder chains triggers and action steps, agents are callable as steps inside automations, and automations can be handed to agents as tools, which the vendor documents as the way to build multi agent workflows with deterministic control; synchronous automations return results to callers. Budibase agents launch post, pricing comparison table and agent tools docs, 2 Oct 2026.

Sourcebudibase.com/blog/updates/ai-agents-betaread 2026-10-02

A maintained retrieval structure: knowledge bases are configured on a pgvector database, assigned to an agent operation with a chosen embedding model, tuned by minimum similarity and chunks to retrieve, and filled with uploaded files and knowledge content that persist for semantic retrieval. pgvector is the only supported provider. Budibase agent vector databases docs and pricing table (Knowledge Sources), 2 Oct 2026.

Sourcedocs.budibase.com/docs/agent-vector-databases.mdread 2026-10-02

Approval before an action commits ships with the product: each tool can carry an unconditional or conditional approval rule (for example only when an input field exceeds a threshold), tied to a reusable policy naming the approvers and whether any one, a majority or all must agree, with approval cards in Slack or Microsoft Teams showing selected parameters and the requester; automation triggered agents pause and resume after approval. Tools also run as the requester or an elevated admin. Budibase agent tools and permissions docs, 2 Oct 2026.

Sourcedocs.budibase.com/docs/agent-tools-and-permissions.mdread 2026-10-02

Budibase documents both access controls and a certification. Access: role based access control with tenant and app roles, user groups, SSO with PKCE OIDC and enforceable SSO, SCIM directory sync on Enterprise, and agent tools that run under the requester's permissions with metadata and results redacted when the requester lacks access. Compliance: the site footer states ISO 27001 certified and GDPR compliant, with a security page linked from the navigation. Audit logs are available on Enterprise. Budibase pricing comparison table and footer, agent tools docs, 2 Oct 2026.

Sourcebudibase.com/pricingread 2026-10-02

Runtime record of agent work: an Activity timeline records agent tool calls and detailed failure information, agent logs are retained from 1 day on Pro to unlimited on Enterprise and the open source edition, automation runs are inspectable in a logs view, and approval cards name the requester and tool. Platform audit logs sit on Enterprise. The limit: the depth of reasoning traces is not described. Budibase agent tools docs and pricing comparison table, 2 Oct 2026.

Sourcedocs.budibase.com/docs/agent-tools-and-permissions.mdread 2026-10-02

Budibase keeps conversation state within a session: Slack threads act as conversation boundaries with an optional idle timeout, and preview chat history persists in browser session storage for testing. No memory layer with a stated scope and lifetime across conversations is documented; workspace tables the agent reads are the application's data model. Budibase Slack messaging channel docs and agent testing guide, 2 Oct 2026.

Sourcedocs.budibase.com/docs/messaging-channels-slackread 2026-10-02

Named customer environment options: the open source edition and Enterprise self host on the customer's own infrastructure or on premises, Business adds a self host on premise option, and air gapped deployment with an offline license is an Enterprise add on; Budibase Cloud is the managed alternative. Budibase pricing page and FAQ, 2 Oct 2026.

Sourcebudibase.com/pricingread 2026-10-02

The customer assembles its own agents: the agent builder supplies a structured instruction template, and the docs walk through building a service desk agent step by step. No catalog of prebuilt, installable agents appears on the pricing page, product navigation or agent docs index. Budibase agent building 101 and docs AI index, 2 Oct 2026.

Sourcedocs.budibase.com/docs/agent-building-101read 2026-10-02

Agents run autonomously through automations, which start from 20+ built in triggers including webhooks and data events and can call an agent as a step; agents triggered this way run and pause on escalation without a person starting them. Agents are also deployed to Slack, Microsoft Teams and Discord through messaging channels, with Slack threads as conversation boundaries. Budibase pricing comparison table, agent tools docs and Slack messaging channel docs, 2 Oct 2026.

Sourcebudibase.com/pricingread 2026-10-02

The customer chooses: Agents Config connects the customer's own model providers and credentials (bring your own AI key on every plan), sets a default model, and attaches a model per agent; embedding models for knowledge retrieval are chosen separately. Budibase AI credits are an alternative on Cloud. Budibase docs AI section and pricing comparison table, 2 Oct 2026.

Sourcedocs.budibase.com/docs/airead 2026-10-02

A documented API for the vendor's own platform: the public REST API works with apps, users, tables and data, with an expanded API on higher tiers, endpoints published in OpenAPI through the docs llms.txt, webhook automations that outside systems call, a CLI, and UI, datasource and automation plugins. Budibase pricing comparison table and docs llms.txt, 2 Oct 2026.

Sourcebudibase.com/pricingread 2026-10-02

Budibase documents a testing method rather than a harness: the pricing page lists Evals and the testing guide tells the builder to assemble a prompt set, define pass criteria and rerun it after any prompt or tool change, using a preview chat that simulates roles through Test as and offers mock approve and reject on escalations. Nothing stores, scores or gates the test cases, and the regression comparison is manual. Budibase agent testing guide and pricing comparison table, 2 Oct 2026.

Sourcedocs.budibase.com/docs/agent-testing-guide.mdread 2026-10-02

No browser, desktop or remote computer control by the agent is documented; agent tools are tables, REST requests, integrated apps, knowledge sources and automations. Budibase agent tools and permissions docs and docs AI index, 2 Oct 2026.

Sourcedocs.budibase.com/docs/agent-tools-and-permissions.mdread 2026-10-02