Assembled vs Thread (2026)

Native integrations with Zendesk, Salesforce, Shopify and others pull data and trigger actions automatically, and interactions are enriched with order history and account details drawn from those systems. Agents carry out actions at the account level instead of only answering, and Copilot offers task execution in one click inside the customer's existing ticketing and CRM. A public API at docs.assembled.com lets customers embed agents in their own product, and connectors that install in one click are listed on the integrations page.

Sourceassembled.com/features/agentic-workflowsread 2026-09-05

Native connections reach the systems an MSP runs on, with write actions into the ticket record. Thread integrates with the three main MSP PSAs, ConnectWise PSA, Autotask and HaloPSA, plus Microsoft Teams, Slack and Microsoft 365, and the agents write into them as well as reading.

Triage sets ticket properties, categorizes, prioritizes, assigns a technician and creates time entries automatically, and Thread reports 96% accuracy on those properties.

Beyond the PSA, Thread connects to Rewst, Power Automate and Zapier for automation and to Pia for full ticket resolution, with Triage passing structured data into Pia to complete PSA and Microsoft 365 actions end to end.

Sourcegetthread.comread 2026-09-05

Specialized agents do the work in several steps, instead of one model doing everything.

Agentic workflows use several specialized AI agents to handle multi step, cross channel journeys end to end, combining deterministic logic the customer can rely on with AI reasoning for the unpredictable parts, and Assembled says each agent holds one task so accuracy holds across the chain.

Above the individual case, support orchestration ties automation to staffing, scoring which case types to automate and rebalancing in real time with dynamic handoff logic across human agents, AI agents and outsourcing partners.

Sourceassembled.com/features/agentic-workflowsread 2026-09-05

Several specialized agents cover one journey end to end, with structured handoff between them and out to another platform. The System of Agents divides the service desk by job.

Triage identifies the request, gathers context and triggers the workflow, Overflow catches calls that would go to voicemail and escalates a P1 to the on call team, Reminder checks whether an issue is still live and closes threads that resolved themselves, and Memory maintains the contact record underneath.

Triage passes structured data into Pia to complete PSA and Microsoft 365 automation, so a routine password or access request finishes without a technician acting as middleman, and everything else routes to the right person with the ticket already categorized, prioritized and prepped.

Sourcegetthread.com/blog/magic-agents-a-new-era-of-ai-powered-automation-for-mspsread 2026-09-05

Work reaches the agent by phone, chat and email, and the agent receives work instead of starting it. AI Voice, AI Chat and AI Copilot are named agent products alongside AI Email, and agentic workflows handle cross channel journeys. There is no outbound or proactive trigger, so the agent answers what arrives instead of starting a conversation. SMS is not among the named agents, which are Voice, Chat and Email only. Real time handoff logic and SLA alerts responding to queue demand are workforce management signals inside the customer's own operation, not triggers reaching the agent.

Sourceassembled.com/features/ai-voice-agentread 2026-09-05

Five channels include the two an MSP desk lives in, and work starts three ways. Agents chat, email and call across chat, email, phone, Microsoft Teams and Slack, and Teams and Slack matter because that is where an MSP's end users raise issues, not a help portal. Thread Voice AI is generally available and automates the phone channel with capture, transcription and automatic ticketing.

Work starts inbound, when a client messages or calls, on events, when the Overflow Agent picks up a call that would otherwise reach voicemail and escalates a P1 to the on call team outside hours, and proactively, when the Reminder Agent reaches back out to check whether an open issue is still live and closes the thread if it resolved itself. Traditional phone and email stay fully active alongside.

Sourcegetthread.comread 2026-09-05

A shared knowledge layer exists, but its internals are not described. Copilot serves agents policies, procedures and knowledge without their having to search documentation, and Assembled says Copilot shares the same knowledge, workflows and data as the AI agents and the workforce management side, so one corpus serves the whole operation.

Interactions are enriched with customer context pulled from connected systems. There is no description of an index, embeddings layer, retrieval architecture, refresh behavior or citation to sources, and no knowledge product page. Knowledge gap analysis is a report on automation opportunities, not a way to ground answers.

Sourceassembled.com/features/ai-copilotread 2026-09-05

Thread grounds its agents in ticket resolution history, knowledge articles and ITIL best practice, and builds a living service catalog from historical tickets, so categories and answers reflect how that MSP actually works instead of a generic taxonomy, and the catalog updates as tickets accumulate. Thread describes the underlying conversational dataset as the largest in the MSP market, drawn from more than 170 million tickets across 750-plus partners. Device, preference and issue history for each contact sits with the Memory Agent.

Sourcegetthread.comread 2026-09-05

There is no memory layer with a stated scope or lifetime. What exists is case continuity and retrieval. Copilot gives agents full case history and keeps context consistent when a case shifts teams, channels or time zones, and interactions are enriched with order history, account details and preferences pulled from the customer's connected systems at the moment of the interaction. Both read records the customer's own helpdesk and commerce systems hold, which is retrieval, not state the agent retains. There is no persistence a customer could name, scope or delete, and no memory product.

Sourceassembled.com/features/ai-copilotread 2026-09-05

The Memory Agent, part of Service Intelligence, creates a living memory for every contact, remembering their devices, preferences and past issues and updating automatically as new interactions happen, so any technician on the desk can serve a client as if they already know them.

The platform forms this state from interactions and keeps it across separate conversations, tied to an identifiable contact, beyond a lookup against the PSA. MSP technicians rotate, and the memory is what carries over when they do. Ticket resolution history, ITIL practice and the service catalog built from historical tickets sit on the knowledge side.

Sourcegetthread.com/blog/magic-agents-a-new-era-of-ai-powered-automation-for-mspsread 2026-09-05

The customer defines the gate, and it sits inside the flow. The customer sets the exact guardrails and policies, including steps for human review or escalation placed within the workflow itself, so a sensitive path stops for a person instead of being routed after the fact. Agentic workflows combine LLM reasoning with hard rules that enforce policy every time, giving deterministic control where it is needed, and brand style and company guidelines are configurable so responses hold to a set standard. Copilot carries the human side, guiding agents through cases that cannot be fully automated.

Sourceassembled.com/blog/agentic-workflowsread 2026-09-05

The partner sets the autonomy boundary, and the end user always has a way out. Magic Agents are customized to the security and permissions each MSP needs, so the partner defines what agents may do in their clients' environments instead of accepting a default. Autonomy is bounded on purpose, with agents resolving 15 to 25 percent of tickets end to end and handing the rest to a technician fully prepped.

Thread says the AI's job is triage and information gathering and that anything complex hands off immediately with full context, and end users are told that typing "speak to an engineer" at any point bypasses the assistant entirely. Automations built on Rewst, Power Automate and Zapier are described as human in the loop. There is no approval queue before individual agent actions go out.

Sourcedocs.getthread.com/article/i6mm06bsv7-thread-partner-change-management-guideread 2026-09-05

SOC 2 Type II is attested by The Cadence Group against the 2017 Trust Services Criteria, and Assembled is HIPAA compliant, with Business Associate Agreements in place with every entity that could transfer protected health information and a BAA available on request.

It adheres to GDPR and has renewed its certification under the EU-U.S. Data Privacy Framework, holds an SSL Labs A+ rating, runs annual third party penetration testing using dynamic fault injection and source code review, checks security across the development lifecycle, trains employees on security every year and runs a named incident response channel. A Vanta hosted trust center at trust.assembled.com holds SOC 2 reports, penetration test reports and incident response plans, gated and opened by hand on request.

For access, the platform offers role based access control, multi factor authentication and single sign on, which customers use themselves, with AES-256 at rest and SSL/TLS in transit, and employee access to customer data is logged and audited. ISO 27001 is not claimed.

Sourcesupport.assembled.com/hc/en-us/articles/40487316559629-how-does-assembled-keep-data-secureread 2026-09-05

The security program is in place, but the SOC 2 attestation has not been issued yet. The Trust Center publishes continuous control monitoring across 107 controls, a subprocessor list, public data handling documents and restricted reports available on request.

Thread acts as a data processor with processing terms available on request, data is encrypted, communications are logged into the ticketing system, and Thread states customer data is not used to train Claude or any other model, not by AWS, not by Anthropic and not by Thread. The SOC 2 Type II audit is in progress, with a bridge letter available under NDA, and Thread tells partners not to describe it as SOC 2 certified until the report is issued. AWS and Azure hold SOC 2 Type II and ISO 27001 for the clouds Thread runs on, not for Thread itself.

Sourcedocs.getthread.com/security-billing/soc-2-and-compliance-statusread 2026-09-05

Every interaction has a complete decision log covering which data was used, what steps were taken and why each path was selected, which Assembled presents as full transparency into AI reasoning. Every action and API call is logged and reviewable, and AI steps are explainable and traceable. Around that sit adoption, usage and impact metrics spanning agent performance, response quality and handle time, plus real time visibility across human agents, AI agents and outsourcing partners.

Sourceassembled.com/features/agentic-workflowsread 2026-09-05

Performance is measured, but execution is not traced. Thread reports ticket properties set at 96% accuracy, sentiment at 97% positive, resolution rates, hours returned to technicians and an ROI calculator, giving a desk manager a clear read on what the agents achieved across more than 170 million tickets.

Communications are logged into the ticketing system, so a conversation can be retrieved as a ticket record. There is no reasoning trace for each decision, tool call log, source attribution on an answer or configuration audit trail for an agent, so a manager can see that triage was 96% accurate without being able to reconstruct why a particular ticket was categorized as it was.

Sourcegetthread.comread 2026-09-05

Hosting region and residency options are not stated. Assembled's security information covers encryption, access, penetration testing, incident response and secure development but names no region or cloud provider, and its security and compliance help article answers SOC 2, HIPAA and GDPR but not hosting location.

The trust center at trust.assembled.com is gated, with access granted by hand on request within a few days, and holds SOC 2 reports, penetration test reports and incident response plans, not a residency offering. The EU-U.S. Data Privacy Framework re-certification covers data transfer, not a location a customer selects.

Thread names one region and offers no choice. Data is held in the United States, application data lives in AWS US East in Northern Virginia, and Claude inference runs on Bedrock's US region capacity, with Azure carrying the other AI features. There is no alternative region, no location the customer selects, and no self hosted, private cloud or single tenant option. Thread acts as a data processor, with processing terms available on request and a subprocessor list published at the Trust Center.

Sourcedocs.getthread.com/article/2gttcwtp4a-thread-ai-on-aws-bedrock-anthropic-claude-data-processing-security-triage-agent-super-magicread 2026-09-05

Customers adopt a named set of agents by role and channel, each with its own product page, with AI Voice, AI Chat, AI Email and AI Copilot on the agent side and AI powered scheduling, ML based forecasting, real time analytics and vendor management as workforce modules. A customer picks the ones its operation needs instead of assembling each from parts. Custom agents are created from scratch, not cloned, and there is no browsable template gallery or marketplace, so customers adopt from the named product set.

Sourceassembled.com/features/ai-copilotread 2026-09-05

A named agent set ships for an MSP to switch on, not a catalog to browse. The System of Agents has a Triage Agent that collects context, categorizes and routes, a Reminder Agent that checks whether issues are still active and closes resolved threads, an Overflow Agent that stops calls reaching voicemail, and a Memory Agent inside Service Intelligence.

Each works out of the box without a partner building it. The four agents are a product set, not a library a customer browses, and there is no template gallery. A community driven marketplace of service workflows and first party automations is on Thread's roadmap and not available today.

Sourcegetthread.com/blog/magic-agents-a-new-era-of-ai-powered-automation-for-mspsread 2026-09-05

Customers get no model selection, routing or bring your own model control. Assembled promotes a webinar on building a Claude skill, and a model provider's customer story describes Assembled Assist running on Claude and switching providers within twenty minutes during another provider's outage, which points to a third party frontier model behind the product that Assembled can change at its discretion. What is missing is customer choice, not flexibility on Assembled's side.

Sourceassembled.com/securityread 2026-09-05

Two model platforms are in use and Thread chooses between them. Claude runs inside AWS Bedrock for the Triage Agent and Super Magic features, Thread's other AI features run on Microsoft Azure, and these features are being turned on progressively. The platform runs across at least two providers, Anthropic through Bedrock and Azure OpenAI, and has moved a flagship agent from one to the other, but the selection sits with Thread. There is no model selector, routing policy or bring your own key path for customers, and a customer cannot choose which model handles their tickets.

Sourcedocs.getthread.com/article/2gttcwtp4a-thread-ai-on-aws-bedrock-anthropic-claude-data-processing-security-triage-agent-super-magicread 2026-09-05

A public interface lets outside systems call the platform. API documentation at docs.assembled.com is linked from the site navigation and the footer, so a developer can read it without a sales conversation or an account. An integrations surface sits alongside it.

Sourcedocs.assembled.com/#introductionread 2026-09-05

There is no public API, SDK, webhook reference or MCP endpoint. Partners extend Thread through other platforms, with automation through Rewst, Power Automate and Zapier and full ticket resolution through Pia, so they connect a third party automation tool instead of calling Thread directly. A community driven marketplace of service workflows is on Thread's roadmap and not available today.

Sourcegetthread.com/blog/magic-agents-a-new-era-of-ai-powered-automation-for-mspsread 2026-09-05

Measurement happens after deployment, with no gate before it. Assembled tracks adoption, usage and impact across agent performance, response quality and handle time, reports automation impact, and scores which case types to automate, so a change can be judged against readable outcomes once live. There is no simulation, scenario set, regression suite or scored evaluation of a proposed agent change before customers see it. Scenario testing of schedules and forecast accuracy reporting exercise staffing plans and forecasts, not the agent.

Sourceassembled.com/features/ai-copilotread 2026-09-05

Outcomes are measured and backed by a commercial guarantee, but nothing tests a change before it reaches clients. Thread tracks triage accuracy at 96%, sentiment at 97% positive, resolution rates and hours returned, and backs it with a 60 day positive ROI guarantee or a refund. There is no simulation, staging environment, scenario suite, regression run or scored evaluation of an agent before release, so tuning happens against live client tickets instead of in a sandbox.

Sourcegetthread.comread 2026-09-05

Agents act programmatically, not through a browser, hosted session, virtual desktop or computer use. They work through native integrations with Zendesk, Salesforce, Shopify and others that pull data and trigger actions, through actions at the account level inside connected systems, and through a public API.

Sourceassembled.com/features/agentic-workflowsread 2026-09-05

There is no browser control, hosted session or interface automation. Agents work through integrations, writing into ConnectWise, Autotask and HaloPSA through native connections, handing structured data to Pia to complete PSA and Microsoft 365 actions, and reaching people over chat, email, Teams, Slack and phone. Where an action falls outside those connectors, the path is a third party automation platform such as Rewst, Power Automate or Zapier, not an agent operating a screen.

Sourcegetthread.comread 2026-09-05