Inkeep Agents
Build synchronized AI assistants and multi-agent workflows in a visual builder or TypeScript SDK.
Per-dimension scores and reasoning
Evidence: The repository contains extensive dependency overrides (pnpm overrides in package.json), indicating some attention to dependency security; README mentions MCP tools and credential management, suggesting sensitive data handling mechanisms; CI workflows (e.g., auto-format) and version control (changesets) support rollback. However, there is no user confirmation mechanism, detailed documentation of data flow transparency, or explicit least-privilege principle. Deductions: No user confirmation or data flow transparency implementation/documentation; sensitive data handling only mentions credential management without depth; external effects (e.g., MCP tool calls) lack security boundary descriptions; rollback relies only on version control, no data rollback plan; source attribution only via author field, unverified.
Evidence: Repository structure is clear, multiple packages (agents-api, agents-sdk, etc.) have clear responsibilities, self-consistency is good; dependencies are locked via pnpm-lock.yaml, and many overrides fix known vulnerabilities, so dependency availability is good; however, failure messages lack detailed error handling or user-friendly failure prompts in README and code. Deductions: Failure messages lack concrete examples or documentation; self-consistency is good but not perfect due to possibly incomplete docs.
Evidence: README clearly lists multiple use cases (customer experience agents, internal copilot, workflow automation), broad audience; environment fit: supports Vercel and Docker deployment, compatible with Vercel AI SDK; but capability boundaries and trigger precision are not detailed. Deductions: Capability boundaries not explicit, trigger precision only mentions MCP, A2A, etc., without depth.
Evidence: Information architecture is clear, README provides architecture overview and component descriptions; install notes have quick start link but lack detailed steps; naming stability: package names and versions exist, but naming conventions not explained; examples and FAQ: code examples present, but FAQ missing; known limitations not listed; license is ELv2 with supplemental terms, but metadata is NOASSERTION; versioning and changelog: changesets configured, but no CHANGELOG file provided; maintenance responsibility via author and community links, but not explicit. Deductions: Install notes not detailed, naming stability not explained, known limitations missing, version control exists but no actual changelog shown, maintenance responsibility unclear.
Evidence: Output usability: provides UI component library and SDK, output directly usable for building chat interfaces; marginal value: offers both no-code and code approaches, differentiated value; cost-benefit: open-source free, but deployment and maintenance costs not assessed. Deductions: Cost-benefit lacks concrete analysis, marginal value good but not quantified.
Evidence: README claims (e.g., 2-way sync) are supported by code and docs, but no test results or independent verification; cross-source corroboration: only relies on internal repo docs, no external validation; fact-inference separation: README distinguishes feature descriptions and architecture, but some claims may be inferred. Deductions: Lack of test evidence and external corroboration, fact-inference separation not strict.
- Repository metadata license is NOASSERTION, actual license is ELv2 with supplemental terms, need to confirm compliance.
- No user confirmation mechanism, agents may automatically execute external actions, deploy with caution.
- Sensitive data handling only mentions credential management, no details on encryption and storage.
- Dependency security has overrides but no vulnerability scan report or security audit.
- Rollback relies only on version control, no data rollback or migration rollback plan.
What does this agent do, and when should you use it?
Inkeep Agents is a platform for AI assistants and multi-agent workflows, offering both a no-code Visual Builder and a TypeScript Agents SDK with two-way synchronization. Its components include agents-api for configuration, execution, and evaluation; the drag-and-drop agents-manage-ui; the declarative @inkeep/agents-sdk; and agents-cli for synchronization. The API manages Agents, Sub Agents, MCP Servers, Credentials, and Projects, tracks conversation state, and emits OpenTelemetry traces. It also includes agents-ui for embedded chat experiences and supports triggering through MCP, A2A, and Vercel SDK APIs, with deployment described for Vercel or Docker. Under the hood it uses the Vercel AI SDK for LLM-provider integration and is compatible with Vercel's useChat hook.
Teams can create and manage Agents on the drag-and-drop agents-manage-ui canvas, or declare Agents, subagents, prompts, and available MCP tools in TypeScript with agent and subAgent from @inkeep/agents-sdk. Both the SDK and Visual Builder write to agents-api; that API stores configuration for Agents, Sub Agents, MCP Servers, Credentials, and Projects, and exposes Agent execution and evaluation. During execution, it tracks conversation state and emits OTEL traces. agents-cli provides inkeep push and inkeep pull to synchronize TypeScript SDK code with the Visual Builder, while agents-ui embeds dynamic chat interfaces in web applications.
- A help-center owner builds a real-time customer assistant in the Visual Builder while engineers continue maintaining it through the SDK.
- A documentation team and engineering team co-own a product-docs assistant: nontechnical contributors adjust the canvas while developers define subagents in TypeScript.
- A support-operations team creates automation for CRM updates and helpdesk-ticket triage, using MCP tools for the required connections.
- An internal platform team deploys copilots for sales, marketing, operations, or support and embeds the chat experience with agents-ui.
- An engineering team coordinating several specialist subagents uses an Agent's defaultSubAgent and subAgents configuration to define that structure.
What are this agent's strengths and limitations?
- The Visual Builder and TypeScript SDK are described as two-way synchronized, allowing nontechnical and engineering teams to collaborate on the same Agent definitions.
- agents-api brings configuration, execution, evaluation, conversation state, and OTEL traces into one service boundary.
- The platform explicitly combines multi-agent architecture, MCP tools with credential management, and agents-ui for embeddable chat interfaces.
- It documents MCP, A2A, and Vercel SDK API triggering paths, plus Vercel and Docker deployment options.
- The supplied material lacks installation commands, runtime versions, environment variables, authentication details, and end-to-end deployment configuration, creating an onboarding gap.
- Its LLM integration is built on the Vercel AI SDK; although it says users can choose an LLM provider, the material does not enumerate providers, adapters, or a support matrix.
- MCP tool use includes credential management, but the material does not specify the permission model, secret storage, or failure-handling behavior.
- Repository metadata says NOASSERTION while the README describes ELv2 subject to Supplemental Terms; adopters should verify the applicable terms.
How do you install or deploy this agent?
The supplied material does not provide copyable installation commands, a Node.js version, environment variables, credential setup, or Docker/Vercel deployment configuration, so reliable installation instructions cannot be derived. It only states that deployment can use Vercel or Docker and names the agents-cli synchronization commands inkeep push and inkeep pull.
How do you use this agent?
The documented minimal code pattern imports agent and subAgent from @inkeep/agents-sdk, uses subAgent to define an id, name, description, canUse, and prompt, then uses agent to define the top-level Agent with id, name, description, defaultSubAgent, and subAgents. inkeep push or inkeep pull can then synchronize code and the Visual Builder. The supplied material does not include package-installation, authentication, or first Agent-execution command/API details; those are the missing pieces for a runnable first invocation.