LOOPmKit and the others: a comparison with sources
Nine products read on their own pages, twenty-one rows, and the date we checked. Unknown means the pages do not say; it does not mean no.
- Found
- Partly
- No
- Unknown: the pages read do not say
- Taken from a search summary, not verified on the vendor page
Checked on 4 November 2026. The LOOPmKit column comes from our own feature inventory and marks what is built but not rolled out, or tested only against stand-ins.
What we did not find elsewhere
Automatic routing to the cheapest model that fits.
Built, with a cascade that learns from past attempts.
Three levels of trust for every action: read, draft, act.
Built. Approving each dynamic call is not supported yet.
Not found in the public docs we read, 9 products, checked 4 November 2026.
Where the others lead
A hosted cloud, customers, certifications, a marketplace and a community, the number of connectors, telephony, and an open licence. We have none of these today, or only in part.
The full table
| Row | LOOPmKit | n8n | Zapier Agents | Dify | Flowise | CrewAI (OSS + AMP) | LangGraph / LangSmith Deployment | Relevance AI | Lindy | Microsoft Copilot Studio |
|---|---|---|---|---|---|---|---|---|---|---|
| Self-hosting and license | ||||||||||
| Visual canvas | ||||||||||
| Agents trigger each other through records and events† | ||||||||||
| Model providers, local models included† | ||||||||||
| Automatic routing to the cheapest suitable model | ||||||||||
| Hard budgets per run, agent and project that stop a run | ||||||||||
| Built-in savings: deduplication, response and prompt cache | ||||||||||
| Approval queue with editing of arguments | ||||||||||
| Trust levels: read, draft, act† | ||||||||||
| Built-in persistent memory† | ||||||||||
| Evals and golden datasets† | ||||||||||
| MCP as client and as server† | ||||||||||
| Browser and desktop control† | ||||||||||
| Isolated code execution† | ||||||||||
| Company wizard with roles† | ||||||||||
| Number of connectors† | ||||||||||
| Telephony† | ||||||||||
| Hosted cloud offer | ||||||||||
| Ready marketplace and community size | ||||||||||
| SOC 2 or other enterprise certification | ||||||||||
| Customers |
-
Self-hosting and license
Self-host Community edition. Sustainable Use License (fair-code, not OSI open source) plus a separate Enterprise License. Source
-
Visual canvas
Home page: 'a visual builder and drop into code whenever you need it'. Source
-
Agents trigger each other through records and events
LOOPmKit
n8n
No record-event agent chaining doc was checked; other forms of chaining may exist.
-
Model providers, local models included
Ollama Chat Model node; OpenAI, Anthropic, Google and open-source providers. Source
-
Automatic routing to the cheapest suitable model
LOOPmKit
n8n
Not found in the vendor docs read.
-
Hard budgets per run, agent and project that stop a run
LOOPmKit
n8n
No budget control found on pages read.
-
Built-in savings: deduplication, response and prompt cache
LOOPmKit
n8n
Not stated on the vendor pages read; not assumed to be absent.
-
Approval queue with editing of arguments
Human-in-the-loop for tool calls via Slack, Telegram or Chat; docs describe approve or deny only, no argument editing. Source
-
Trust levels: read, draft, act
LOOPmKit
n8n
No read/draft/act ladder found in the docs read.
-
Built-in persistent memory
Docs link a 'How memory works' page; details not read. Source
-
Evals and golden datasets
Evaluations with test datasets and metrics; plan availability not stated. Source
-
MCP as client and as server
MCP Client Tool node and MCP Server Trigger node. Source
-
Browser and desktop control
LOOPmKit
n8n
Not stated on the vendor pages read; not assumed to be absent.
-
Isolated code execution
LOOPmKit
n8n
Not checked on the pages read.
-
Company wizard with roles
LOOPmKit
n8n
Not stated on the vendor pages read; not assumed to be absent.
-
Number of connectors
500+ native integrations on home page; README says 1500+ while repo About says 400+ (inconsistent). Source
-
Telephony
LOOPmKit
n8n
Not stated on the vendor pages read; not assumed to be absent.
-
Hosted cloud offer
n8n Cloud Starter from 20 EUR per month. Source
-
Ready marketplace and community size
209.5k GitHub stars; forum 'over 170,000' members. Source
-
SOC 2 or other enterprise certification
Trust Center lists SOC 2 Type 2 and SOC 3. Source
-
Customers
Case studies Huel and Vodafone; 'Trusted by Fortune 500 companies' (claim). Source
-
Self-hosting and license
LOOPmKit
Zapier Agents
Not stated on the vendor pages read; not assumed to be absent.
-
Visual canvas
LOOPmKit
Zapier Agents
Agent builder UI not described as a canvas on the pages read.
-
Agents trigger each other through records and events
LOOPmKit
Zapier Agents
No record-event agent chaining doc was checked; other forms of chaining may exist.
-
Model providers, local models included
LOOPmKit
Zapier Agents
Not stated on the vendor pages read; not assumed to be absent.
-
Automatic routing to the cheapest suitable model
LOOPmKit
Zapier Agents
Not found in the vendor docs read.
-
Hard budgets per run, agent and project that stop a run
Per-run activity cap (10 free, 40 paid); at the cap the agent asks before continuing. No per-project money budget found. Source
-
Built-in savings: deduplication, response and prompt cache
LOOPmKit
Zapier Agents
Not stated on the vendor pages read; not assumed to be absent.
-
Approval queue with editing of arguments
Human in the Loop for Zaps lets reviewers approve, decline or change data; not confirmed for Agents. Source
-
Trust levels: read, draft, act
LOOPmKit
Zapier Agents
No read/draft/act ladder found in the docs read.
-
Built-in persistent memory
LOOPmKit
Zapier Agents
Not stated on the vendor pages read; not assumed to be absent.
-
Evals and golden datasets
LOOPmKit
Zapier Agents
Not stated on the vendor pages read; not assumed to be absent.
-
MCP as client and as server
Zapier MCP lets AI clients use Zapier apps (server side); Agents as MCP client not confirmed. Source
-
Browser and desktop control
Agents can browse and scrape the web and use a Chrome extension; desktop control not stated. Source
-
Isolated code execution
LOOPmKit
Zapier Agents
Not checked on the pages read.
-
Company wizard with roles
LOOPmKit
Zapier Agents
Not stated on the vendor pages read; not assumed to be absent.
-
Number of connectors
9,000+ apps. Source
-
Telephony
LOOPmKit
Zapier Agents
Not stated on the vendor pages read; not assumed to be absent.
-
Hosted cloud offer
Hosted SaaS; Agents add-on free tier 400 activities per month. Source
-
Ready marketplace and community size
9,000+ app ecosystem. Source
-
SOC 2 or other enterprise certification
SOC 2 Type II and SOC 3 reported by Zapier trust pages (via search summary; trust page itself not fetched). Source
-
Customers
Claims 2.2 million to 3.4 million companies (inconsistent figures on the same page). Source
-
Self-hosting and license
Community Edition via Docker Compose. Dify Open Source License, described as Apache 2.0 with additional conditions; Community tier aimed at non-commercial use. Source
-
Visual canvas
Visual canvas for workflows and agents. Source
-
Agents trigger each other through records and events
LOOPmKit
Dify
No record-event agent chaining doc was checked; other forms of chaining may exist.
-
Model providers, local models included
Hundreds of models and any OpenAI-compatible API; local models not explicitly stated on the page read. Source
-
Automatic routing to the cheapest suitable model
LOOPmKit
Dify
Not found in the vendor docs read.
-
Hard budgets per run, agent and project that stop a run
LOOPmKit
Dify
Plans have message-credit quotas; not a per-run or per-agent budget.
-
Built-in savings: deduplication, response and prompt cache
LOOPmKit
Dify
Not stated on the vendor pages read; not assumed to be absent.
-
Approval queue with editing of arguments
Human Input node: reviewer edits pre-filled content and chooses a branch; this is content review, not tool-argument editing. Source
-
Trust levels: read, draft, act
LOOPmKit
Dify
No read/draft/act ladder found in the docs read.
-
Built-in persistent memory
LOOPmKit
Dify
Knowledge bases documented; agent memory not confirmed.
-
Evals and golden datasets
LOOPmKit
Dify
Retrieval testing only on the page read.
-
MCP as client and as server
MCP servers as tools (HTTP transport only); publishing as MCP server referenced but not read. Source
-
Browser and desktop control
LOOPmKit
Dify
Not stated on the vendor pages read; not assumed to be absent.
-
Isolated code execution
LOOPmKit
Dify
Not checked on the pages read.
-
Company wizard with roles
LOOPmKit
Dify
Not stated on the vendor pages read; not assumed to be absent.
-
Number of connectors
LOOPmKit
Dify
Marketplace exists; no count read.
-
Telephony
LOOPmKit
Dify
Not stated on the vendor pages read; not assumed to be absent.
-
Hosted cloud offer
Dify Cloud, Sandbox free to Team 1,590 USD per year. Source
-
Ready marketplace and community size
About 160.4k GitHub stars; Dify Marketplace for plugins and templates. Source
-
SOC 2 or other enterprise certification
SOC 2 and ISO 27001 badges on pricing page without audit details. Source
-
Customers
LOOPmKit
Dify
No customers named on pricing page.
-
Self-hosting and license
Apache 2.0, Docker or npm. Repository archived by owner on 2026-08-13. Source
-
Visual canvas
Visual builder for agents and workflows. Source
-
Agents trigger each other through records and events
LOOPmKit
Flowise
No record-event agent chaining doc was checked; other forms of chaining may exist.
-
Model providers, local models included
LOOPmKit
Flowise
README page read does not list providers.
-
Automatic routing to the cheapest suitable model
LOOPmKit
Flowise
Not found in the vendor docs read.
-
Hard budgets per run, agent and project that stop a run
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Built-in savings: deduplication, response and prompt cache
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Approval queue with editing of arguments
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Trust levels: read, draft, act
LOOPmKit
Flowise
No read/draft/act ladder found in the docs read.
-
Built-in persistent memory
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Evals and golden datasets
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
MCP as client and as server
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Browser and desktop control
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Isolated code execution
LOOPmKit
Flowise
Not checked on the pages read.
-
Company wizard with roles
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Number of connectors
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Telephony
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Hosted cloud offer
README mentions Flowise Cloud; repo archived 2026-08-13, current service status not verified. Source
-
Ready marketplace and community size
About 55.5k stars; repository archived. Source
-
SOC 2 or other enterprise certification
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Customers
LOOPmKit
Flowise
Not stated on the vendor pages read; not assumed to be absent.
-
Self-hosting and license
Framework is MIT. AMP Enterprise lists deployment in own VPC or own infrastructure. Source
-
Visual canvas
Crew Studio, visual editor in the free tier. Source
-
Agents trigger each other through records and events
LOOPmKit
CrewAI (OSS + AMP)
No record-event agent chaining doc was checked; other forms of chaining may exist.
-
Model providers, local models included
README: Ollama and LM Studio integration. Source
-
Automatic routing to the cheapest suitable model
LOOPmKit
CrewAI (OSS + AMP)
Not found in the vendor docs read.
-
Hard budgets per run, agent and project that stop a run
LOOPmKit
CrewAI (OSS + AMP)
Not stated on the vendor pages read; not assumed to be absent.
-
Built-in savings: deduplication, response and prompt cache
LOOPmKit
CrewAI (OSS + AMP)
Not stated on the vendor pages read; not assumed to be absent.
-
Approval queue with editing of arguments
README: human-in-the-loop supported; pricing table lists human-in-the-loop input. Source
-
Trust levels: read, draft, act
LOOPmKit
CrewAI (OSS + AMP)
No read/draft/act ladder found in the docs read.
-
Built-in persistent memory
README lists memory among agent capabilities without detail. Source
-
Evals and golden datasets
LOOPmKit
CrewAI (OSS + AMP)
Not stated on the vendor pages read; not assumed to be absent.
-
MCP as client and as server
README cites MCP/A2A support; AMP table lists MCP server export on both plans. Source
-
Browser and desktop control
LOOPmKit
CrewAI (OSS + AMP)
Not stated on the vendor pages read; not assumed to be absent.
-
Isolated code execution
LOOPmKit
CrewAI (OSS + AMP)
Not checked on the pages read.
-
Company wizard with roles
LOOPmKit
CrewAI (OSS + AMP)
Not stated on the vendor pages read; not assumed to be absent.
-
Number of connectors
LOOPmKit
CrewAI (OSS + AMP)
Not stated on the vendor pages read; not assumed to be absent.
-
Telephony
LOOPmKit
CrewAI (OSS + AMP)
Not stated on the vendor pages read; not assumed to be absent.
-
Hosted cloud offer
Free tier with 50 executions per month on CrewAI cloud. Source
-
Ready marketplace and community size
About 60.5k GitHub stars. Source
-
SOC 2 or other enterprise certification
LOOPmKit
CrewAI (OSS + AMP)
Pricing page does not mention SOC 2.
-
Customers
LOOPmKit
CrewAI (OSS + AMP)
Not stated on the vendor pages read; not assumed to be absent.
-
Self-hosting and license
LangGraph library is MIT. Self-hosted platform with control plane needs the Enterprise plan; standalone server needs a LangSmith license. Source
-
Visual canvas
LangSmith Studio for visual prototyping; graphs themselves are defined in code. Source
-
Agents trigger each other through records and events
LOOPmKit
LangGraph / LangSmith Deployment
No record-event agent chaining doc was checked; other forms of chaining may exist.
-
Model providers, local models included
LOOPmKit
LangGraph / LangSmith Deployment
Library page read does not list providers.
-
Automatic routing to the cheapest suitable model
LOOPmKit
LangGraph / LangSmith Deployment
Not found in the vendor docs read.
-
Hard budgets per run, agent and project that stop a run
LOOPmKit
LangGraph / LangSmith Deployment
Not stated on the vendor pages read; not assumed to be absent.
-
Built-in savings: deduplication, response and prompt cache
LOOPmKit
LangGraph / LangSmith Deployment
Not stated on the vendor pages read; not assumed to be absent.
-
Approval queue with editing of arguments
README: inspect and change agent state during execution; approval queue UI not verified. Source
-
Trust levels: read, draft, act
LOOPmKit
LangGraph / LangSmith Deployment
No read/draft/act ladder found in the docs read.
-
Built-in persistent memory
README: long-term persistent memory across sessions and durable execution. Source
-
Evals and golden datasets
LangSmith: dataset collection, online and offline evals on all plans. Source
-
MCP as client and as server
Agent Server exposes each agent at /mcp; client side not verified here. Source
-
Browser and desktop control
LOOPmKit
LangGraph / LangSmith Deployment
Not stated on the vendor pages read; not assumed to be absent.
-
Isolated code execution
LOOPmKit
LangGraph / LangSmith Deployment
Not checked on the pages read.
-
Company wizard with roles
LOOPmKit
LangGraph / LangSmith Deployment
Not stated on the vendor pages read; not assumed to be absent.
-
Number of connectors
LOOPmKit
LangGraph / LangSmith Deployment
Not stated on the vendor pages read; not assumed to be absent.
-
Telephony
LOOPmKit
LangGraph / LangSmith Deployment
Not stated on the vendor pages read; not assumed to be absent.
-
Hosted cloud offer
Cloud in US or EU, Plus 39 USD per seat per month. Source
-
Ready marketplace and community size
About 44.5k GitHub stars. Source
-
SOC 2 or other enterprise certification
LOOPmKit
LangGraph / LangSmith Deployment
Pricing page links a Trust Center, content not read.
-
Customers
LOOPmKit
LangGraph / LangSmith Deployment
Not stated on the vendor pages read; not assumed to be absent.
-
Self-hosting and license
LOOPmKit
Relevance AI
Pricing page does not mention self-hosting or VPC.
-
Visual canvas
LOOPmKit
Relevance AI
Not stated on the vendor pages read; not assumed to be absent.
-
Agents trigger each other through records and events
LOOPmKit
Relevance AI
No record-event agent chaining doc was checked; other forms of chaining may exist.
-
Model providers, local models included
LOOPmKit
Relevance AI
Not stated on the vendor pages read; not assumed to be absent.
-
Automatic routing to the cheapest suitable model
LOOPmKit
Relevance AI
Not found in the vendor docs read.
-
Hard budgets per run, agent and project that stop a run
LOOPmKit
Relevance AI
Not stated on the vendor pages read; not assumed to be absent.
-
Built-in savings: deduplication, response and prompt cache
LOOPmKit
Relevance AI
Not stated on the vendor pages read; not assumed to be absent.
-
Approval queue with editing of arguments
Home page lists 'Human-in-the-loop approvals'; editing not stated. Source
-
Trust levels: read, draft, act
LOOPmKit
Relevance AI
No read/draft/act ladder found in the docs read.
-
Built-in persistent memory
LOOPmKit
Relevance AI
Not stated on the vendor pages read; not assumed to be absent.
-
Evals and golden datasets
LOOPmKit
Relevance AI
Not stated on the vendor pages read; not assumed to be absent.
-
MCP as client and as server
LOOPmKit
Relevance AI
Not stated on the vendor pages read; not assumed to be absent.
-
Browser and desktop control
LOOPmKit
Relevance AI
Not stated on the vendor pages read; not assumed to be absent.
-
Isolated code execution
LOOPmKit
Relevance AI
Not checked on the pages read.
-
Company wizard with roles
Positions itself as an AI Workforce with teams of agents; no role wizard confirmed. Source
-
Number of connectors
2,000+ on pricing page; 1,000+ on home page (inconsistent). Source
-
Telephony
Phone Agent makes outbound calls; Twilio integration; ElevenLabs voice only. Source
-
Hosted cloud offer
Enterprise plan listed on pricing page; hosted product. Source
-
Ready marketplace and community size
LOOPmKit
Relevance AI
Partner directory only; no marketplace stated.
-
SOC 2 or other enterprise certification
SOC 2 Type II badge on home page. Source
-
Customers
Names Canva, Databricks, Confluent, Autodesk and others. Source
-
Self-hosting and license
LOOPmKit
Lindy
Pricing page does not mention self-hosting.
-
Visual canvas
LOOPmKit
Lindy
Not stated on the vendor pages read; not assumed to be absent.
-
Agents trigger each other through records and events
LOOPmKit
Lindy
No record-event agent chaining doc was checked; other forms of chaining may exist.
-
Model providers, local models included
Team plan: choose model per task, most major models; local models not stated. Source
-
Automatic routing to the cheapest suitable model
LOOPmKit
Lindy
Not found in the vendor docs read.
-
Hard budgets per run, agent and project that stop a run
LOOPmKit
Lindy
Not stated on the vendor pages read; not assumed to be absent.
-
Built-in savings: deduplication, response and prompt cache
LOOPmKit
Lindy
Not stated on the vendor pages read; not assumed to be absent.
-
Approval queue with editing of arguments
Write or send actions can wait for approval (vendor marketing); editing not stated. Source
-
Trust levels: read, draft, act
LOOPmKit
Lindy
No read/draft/act ladder found in the docs read.
-
Built-in persistent memory
Home page: memory kept in plain, editable files (vendor marketing). Source
-
Evals and golden datasets
LOOPmKit
Lindy
Not stated on the vendor pages read; not assumed to be absent.
-
MCP as client and as server
Home page says it supports MCP to connect tools; server side not stated. Source
-
Browser and desktop control
LOOPmKit
Lindy
Not stated on the vendor pages read; not assumed to be absent.
-
Isolated code execution
LOOPmKit
Lindy
Not checked on the pages read.
-
Company wizard with roles
Positions agents as AI employees with hireable roles; no company wizard confirmed. Source
-
Number of connectors
1,500+ tools. Source
-
Telephony
Docs: inbound and outbound calls via Make Phone Call action. Source
-
Hosted cloud offer
Free, Team and Enterprise plans. Source
-
Ready marketplace and community size
LOOPmKit
Lindy
Skills and templates mentioned; no size given.
-
SOC 2 or other enterprise certification
SOC 2 Type II, GDPR, HIPAA, PIPEDA listed. Source
-
Customers
Customer logos and testimonials on site; unverified marketing. Source
-
Self-hosting and license
LOOPmKit
Microsoft Copilot Studio
Docs describe it as an Online Service; no self-host option found on pages read.
-
Visual canvas
Graphical interface plus natural language authoring. Source
-
Agents trigger each other through records and events
LOOPmKit
Microsoft Copilot Studio
No record-event agent chaining doc was checked; other forms of chaining may exist.
-
Model providers, local models included
11K+ models via Azure AI Foundry, OpenAI and Anthropic models for computer use; local models not stated. Source
-
Automatic routing to the cheapest suitable model
LOOPmKit
Microsoft Copilot Studio
Not found in the vendor docs read.
-
Hard budgets per run, agent and project that stop a run
Admin can set monthly limit per agent that stops usage, and pay-as-you-go caps; tenant overage blocks agents at 125%. Monthly, not per run. Source
-
Built-in savings: deduplication, response and prompt cache
Prompt-cache read tokens are not billed for reasoning models; no dedupe or response cache described. Source
-
Approval queue with editing of arguments
Human supervision via Outlook for computer-use runs; general tool approval queue not verified. Source
-
Trust levels: read, draft, act
LOOPmKit
Microsoft Copilot Studio
No read/draft/act ladder found in the docs read.
-
Built-in persistent memory
LOOPmKit
Microsoft Copilot Studio
Not stated on the vendor pages read; not assumed to be absent.
-
Evals and golden datasets
LOOPmKit
Microsoft Copilot Studio
Not stated on the vendor pages read; not assumed to be absent.
-
MCP as client and as server
MCP client for tools and resources; building an MCP server is documented as a how-to, prompts not supported. Source
-
Browser and desktop control
Computer use for websites and desktop apps on a Windows machine; Claude and OpenAI computer-use models listed, some experimental. Source
-
Isolated code execution
LOOPmKit
Microsoft Copilot Studio
Not checked on the pages read.
-
Company wizard with roles
LOOPmKit
Microsoft Copilot Studio
Not stated on the vendor pages read; not assumed to be absent.
-
Number of connectors
1,500+ pre-built data connectors. Source
-
Telephony
Voice tiers billed per minute (10, 35, 75 credits). Source
-
Hosted cloud offer
Microsoft-hosted, credit packs or pay-as-you-go. Source
-
Ready marketplace and community size
LOOPmKit
Microsoft Copilot Studio
Connector ecosystem exists; no marketplace figure read.
-
SOC 2 or other enterprise certification
SOC, ISO 27001 and others, HIPAA BAA, HITRUST, FedRAMP, PCI DSS. Source
-
Customers
Named: Estee Lauder, Dow, Amgen, Virgin Money; claims 90% of the Fortune 500 has used it. Source
Checked on 4 November 2026. Notes on other products are quoted in English from the pages read.
Notes on the LOOPmKit column
Where it is built but not rolled out, or tested only against stand-ins.
- Self-hosting and licenseA Docker Compose stack for one operator. The repository is private and has no license, so it is not open source and others cannot self-host it today.
- Visual canvasA canvas of agents and gates.
- Agents trigger each other through records and eventsTriggers on a record, on another agent finishing, on a schedule or a webhook; also an agent as a tool of another. The live call of an agent as a tool awaited a key.
- Model providers, local models included25 provider presets including Ollama, LM Studio and vLLM. The addresses were probed with invalid keys only.
- Automatic routing to the cheapest suitable modelAuto mode picks the cheapest model that meets the tools, vision, context and quality needed, and a learned cascade starts on proven models.
- Hard budgets per run, agent and project that stop a runPer run, per agent per day and per project per day; a run over the cap is skipped, and parallel calls reserve budget first.
- Built-in savings: deduplication, response and prompt cacheThree savers on by default: dedupe by hash, an exact response cache and the provider prompt cache. No semantic cache.
- Approval queue with editing of argumentsDrafts wait in a queue, the approver can edit the arguments, and disclosure is checked again after the edit.
- Trust levels: read, draft, actThe effective level is the lower of the action and the connector scope. Dynamic Gmail calls need act, since per-call approval is not supported yet.
- Built-in persistent memoryA separate memory service from the same author in the same stack; the run goes on if it is down. The "memory required" policy is not rolled out.
- Evals and golden datasets10 kinds of evaluators, golden datasets, replay and comparison of versions and models. The prompt optimiser is not rolled out.
- MCP as client and as serverAs a client, any stdio or HTTP MCP server. As a server, 11 tools; built and not rolled out.
- Browser and desktop controlHeadless Chromium in the API image; a Mac bridge with manual setup; logged-in browser profiles are in progress.
- Isolated code executionPer-job containers with a read-only root, dropped capabilities, limits and an egress allowlist. Component tests only; no run with a real vendor subscription yet.
- Company wizard with roles10 editable roles, up to 32, with budgets and colleague-call rules in one transaction. Checked in a browser with a scripted model. The goal of a company that runs itself is not reached.
- Number of connectors61 presets and any MCP server or REST API. None is checked live one by one.
- TelephonyAn inbound Twilio receptionist, tested against a simulated Twilio and off by default. No real call and no outbound calls.
- Hosted cloud offerSelf-hosting only is documented.
- Ready marketplace and community sizeA private repository and no public community. 23 templates inside the product.
- SOC 2 or other enterprise certificationNo certification.
- CustomersNo external customers and no revenue yet.
Early access starts with a call
The product is at an early stage with a single operator. The repository is private and the terms are not set. If this fits a problem you have, book a short call.
A 20-minute call with the author. The repository is private, so a conversation is the way in.