Hexabot is a workflow automation platform for self-hosted scenarios. The core is to put conversations, actions, memory, RAG, MCP and human review points in the same runtime.

Hexabot

Core parameters and statistics

The main delivery form of Hexabot is Agent/MCP/automation tool, but it is not a lightweight Agent shell that only "calls a few tools", but an AI workflow runtime that is more focused on production systems. It unifies conversation portals, workflows, actions, memory RAGs, MCPs, human review points and multi-channel delivery into a self-hosted platform.

Projects	Public Information
Official positioning	AI Workflows That Talk, Act, and Remember
Product positioning	self-hostable AI workflow automation
Code form	Hexabot v3 source-available
CLI startup	npm install -g @hexabot-ai/cli, hexabot create my-project, hexabot dev
Run entrance	Admin UI http://localhost:3000, API /api, Docs /docs
Core capabilities	workflows, actions, bindings, memory, RAG, MCP, multichannel
Community size	GitHub about 974 stars, 218 forks
Community free version	Community, 1 user, 3 flows
Business Startup	Starter $19/month
Licensing model	Fair Core License, supports self-hosting and expansion

A brief comment: If you want an AI runtime that can actually run business processes, rather than a demonstration chatbot, Hexabot is obviously closer to production than many "Agent Builders".

Promotion Verification: The official promotion of "Talk, Act, and Remember" is basically accurate, because the public capabilities do cover dialogue, actions, memory, and workflow. The core pain point it addresses is not a single question and answer, but "how to orchestrate AI into a business process system that can run sustainably, be self-hosted, and can be taken over manually."

User and market recognition

Hexabot’s recognition comes more from the open source community and the self-hosted workflow crowd than from the purchasing lists of major manufacturers. It is not currently the “hottest” AI product on the Internet, but its positioning is clear for teams that need to control data and execution links.

Community signal: GitHub publicly shows about 974 stars, 218 forks, and 26 contributors. The scale is not explosive, but it is enough to prove that it is not a single-developer toy. The README, CLI, documentation and official website pricing system are relatively complete, indicating that investment in productization is continuous.

Case Signal: The official website publicly mentions Orange's service automation cases. Such cases cannot be directly equivalent to large-scale commercial success, but at least it shows that the team is promoting real customer scenarios rather than just doing open source distribution.

Market Boundary: The official ARR, number of paying customers or retention indicators are not disclosed, so they cannot be overstated. A more accurate judgment is: it has a clear difference in the "self-hosted AI workflow" segment, but whether it is suitable for you depends on whether you really need long-term running, multi-person collaboration and manual review links.

Cost advantage

Hexabot's cost advantage lies in "removing the message volume and the number of integrated connections from the main billing items" and instead using the number of users, workflows and activations as constraints. For teams with stable multi-channel service processes, this model is more predictable than pure usage-based billing.

Plans	Public Prices	Core Limitations	Who Is Suitable for
Community	$0/month	1 user, 3 flows	Learning, validation, self-hosted prototypes
Starter	$19/month	3 activations, 3 users, 25 workflows	Freelancers, small teams, first client projects
Pro	$59/month	10 activations, 10 users, 150 workflows	Enter the team collaboration and sharing management stage
Unlimited	$149/month	25 activations, 25 users, unlimited workflows (fair use)	Resellers, implementation partners, high-usage teams
Enterprise	Customization	Custom quotas, governance and SLAs	High compliance or business-critical scenarios

Free truth: The Community version is essentially a verification layer where “functions can be experienced but production capacity is stuck”. One user with 3 flows is very suitable for proof of concept, but it is completely insufficient for team collaboration and formal operations.

Hidden benefits/costs: It separates the message volume from the core billing items, which is indeed beneficial to budget stability, but self-hosting is not free. You still have to bear server, database, monitoring, backup and upgrade costs. Especially when you access memory, RAG and multi-channel messaging, the operation and maintenance complexity will increase significantly.

Impact on team collaboration: Compared with piecing together multiple Bots, form automation and customer service takeover tools, Hexabot's unified runtime can reduce the rework rate of scattered configurations and unclear responsibilities, but only if the team is willing to accept its workflow and permissions model.

Main functions

• Agentic workflows: Use YAML to define workflows with type constraints, not only for chatting, but also for process status advancement.
• Actions and bindings: Separate action execution and capability binding to facilitate reuse and management.
• Memory, RAG and MCP: Empower workflows with contextual memory, knowledge recall and tool interoperability capabilities.
• Multi-channel delivery: Supports websites, messaging platforms, social channels and custom portals, and delivers the same set of runtimes to multiple service surfaces.
• Human review points: Access human judgment through conditions, inbox workflow and live handoff.

Expert View: The hidden linkage of Hexabot is that "workflow + action + memory + manual review" is designed as a whole. When many teams use a separate Agent framework, it feels fast in the early stage, but starts to fall apart as soon as it is connected to customer service, operations or approval in the later stage; Hexabot is obviously aimed at later management.

Tool open list: What Hexabot exposes to models or workflows is not a separate browser, but a set of orchestration capabilities within the runtime: trigger conversation, run workflow, invoke action, read/write memory, retrieve RAG context, call MCP tool, handoff to human, route to channel, validate schema, store run history. The model uses these capabilities to complete a cycle of "receiving input -> decision-making -> adjusting tools -> writing back status -> switching to manual if necessary".

Model and version evolution

The clearest thing about Hexabot currently is the v3 main line. Products and README are all advancing around v3's runtime, CLI and workflow model.

Latest version: The latest version of the official CLI in the npm registry is 3.2.4, with a release date of 2026-06-12.

Historical node: The previous version 3.2.3 was released on 2026-06-11, showing that the CLI is in a continuous small-step release rhythm. The recent commit of build(release): v3.2.4 can also be seen in the GitHub repository, which is mutually confirmed with the official website "Hexabot V3 is available now".

Version Interpretation: Hexabot version changes are not just interface updates, but will affect CLI initialization, workflow agreements, binding mechanisms and runtime contracts. For teams that want to be self-hosted for a long time, they should first verify whether migration scripts, database compatibility, and custom actions are affected before upgrading.

Technical advantages

The technical advantage of Hexabot is not in the model itself, but in that it advances AI automation from "chat logic" to "process runtime with governance boundaries."

Architecture Link: LLM/Workflow Decision -> Hexabot Runtime -> Actions/Bindings/MCP/Memory/RAG -> External Channels and Business Systems -> Run State/Human Review -> Workflow Continuation

Why it’s more stable: It contracts action input and output, binding capabilities and workflow, instead of allowing the model to play freely. For production scenarios, this is much more reliable than "steps agreed in the prompt words".

Why it saves money: Compared with buying multiple bot platforms, knowledge bases, manual transfers and automation tools separately, Hexabot can unify multi-channel service logic in a self-hosted stack, reducing cross-system back-and-forth synchronization.

Why it is more suitable for business processes: It clearly has built-in human review points, which is very important. Real business AI automation is not 100% unmanned, but knowing where to stop and let people take over.

Guide to engineering pitfalls:

• Dead-end loop and Token inflation control: If the "model judgment + condition return" is designed too loosely in the workflow, it is easy for nodes to call themselves repeatedly. The solution is to set max_steps, an upper limit on the number of loops and a node-level timeout for each workflow run, and block infinite questioning at the runtime layer.
• DOM/Exception context overload: If the full text of external pages, historical messages and knowledge fragments are fully loaded into the model, the RAG hit rate and cost will deteriorate. The solution is to do a hierarchical summary of memory, RAG and external tool returns, and only send the data necessary for this round of decision-making into the context.
• Security and Overreaching Governance: Hexabot supports actions, MCP and multi-channel access, so the risk of overreaching is naturally higher. The solution is to make high-risk actions into actions that require approval, and force irreversible operations such as publishing, deletion, transfer, and outbound calls to human review point.

How to use

Hexabot's entry-level experience is friendlier than most self-hosted Agent frameworks because the official CLI creation path is directly provided.

Entrance	Suitable for the scene	Description
Official CLI	Quick local project start	Best for initial verification
Documentation Station	Learn workflow, extensions and concepts	Suitable for product and development collaboration
GitHub repository	Research source code and custom extensions	Suitable for deep self-hosting
Enterprise solutions	Governance SLA and large-scale rollout	Suitable for formal procurement

Get started quickly in 3 minutes:

npm install -g @hexabot-ai/cli
hexabot create my-project
cd my-project
hexabotdev

Local default entry:

Admin UI: http://localhost:3000
API: http://localhost:3000/api
API docs: http://localhost:3000/docs

Typical approach: Trigger the workflow from a conversation, form, timer or event, call actions, memory, RAG or MCP within the workflow, and then decide to continue automatic execution or transfer to manual review based on the results. This approach is more suitable for the real operation process than "single round of Q&A + webhook".

Product Pricing

Hexabot's pricing clearly serves the self-hosted delivery scenario, rather than treating each message as a revenue stream.

• C client/individual: Community is free and suitable for learning and prototyping.
• Developers/Implementers: Starter and Pro are friendly to freelancers, small teams and implementation partners, with a focus limited to users, workflows and activations.
• Enterprise: Enterprise provides custom quotas, governance, priority support and SLA-backed uptime, which is suitable for high-value service processes.

The advantage of this set of pricing is that it is predictable, but the disadvantage is that when the number of teams, workflows, and deployment nodes increase together, the management complexity will also increase. It eliminates the anxiety of "the more information, the more expensive it is", but it does not eliminate platform governance.

Application scenarios

• Customer Service Automation: Connect the website, social and messaging channels to the same runtime. AI handles high-frequency issues first, and transfers complex issues manually. In terms of working hours, the customer service team's processing time for the first round of repeated inquiries can be reduced from "manual replying one by one" to "only handling exceptions and escalation work orders". The preparation time for a single response can usually be reduced from minutes to tens of seconds, but this is a process deduction, not an official commitment.
• Operation and Growth Process: Use workflow to connect forms, conversations, follow-up actions and CRM to reduce manual copy-paste and cross-tool switching. When junior operations conduct an activity thread diversion, it can often be reduced from 10 to 15 minutes of manual sorting to 2 to 3 minutes for review.
• Internal AI Assistant and Approval Collaboration: Let knowledge retrieval, action execution and manual confirmation points coexist, suitable for teams with compliance needs.

Dimensionality reduction attack scenario: When you have to deal with "multi-channel input + multi-step actions + manual review + self-hosted data boundaries", Hexabot will be much more comfortable than the lightweight Agent Builder.

Not suitable for boundaries: If you only want to build a single-page chatbot, or the team has no operation and maintenance capabilities at all, self-hosted Hexabot will be heavy. It is more suitable for teams with long-term process asset accumulation needs, rather than a one-time activity page tool.

Applicable people

• Self-hosted first product and engineering team: Want to control data, workflow and integration within their own environment.
• Customer service and operation platform team: It is necessary to manage the multi-channel service process in a unified manner and set up manual takeover points.
• Implementation Providers and Resellers: Requires replicable workflow templates, user caps and activation management.

Dissuade the crowd:

• People who just want to experience the chat model: Hexabot is too heavy.
• Small teams without workflow management needs: Introducing actions, bindings, and memory as a set, the learning cost is not low.
• People who cannot afford infrastructure operation and maintenance: Self-hosting is not a slogan, it means that the database, backup, upgrade and security responsibilities are all on your side.

Summary and Outlook

The value of Hexabot is not that it is "yet another Agent framework", but that it puts several things that business processes really need into the same runtime: actions, memories, governance, multi-channel and human confirmation. For serious business scenarios, this is more practical than simply connecting the model to the chat box.

Procurement/adoption risk assessment must also be thoroughly discussed: first, self-hosting will hand over governance rights to you, as well as operation and maintenance responsibilities; second, the Fair Core License and commercial license key solutions mean that you must confirm the authorization boundary as early as possible; third, when it comes to high-value customer delivery, strong compliance approval, and irreversible operations, you cannot be superstitious about full automation, and you must design human-in-the-loop into the main process. Hexabot will only show its advantages when the team really needs long-term operation, multi-person collaboration and auditable execution chain.

Related tools: CrewAI, langchain

Version Info

• Hexabot CLI 3.2.4 ：The latest official CLI version exposed by npm registry, used to create and run Hexabot v3 projects. （2026-06-12）
• Hexabot CLI 3.2.3 ：The previous version of the official CLI shows that v3 is still continuously iterating and polishing the delivery link. （2026-06-11）