Sentra vs Claude Tag: Company Memory Layer vs AI Coworker

TL;DR

• Claude Tag is an AI coworker you @-tag in Slack. It breaks a task into stages, runs autonomously on Opus 4.8, and learns your company from Slack messages.
• Sentra is the memory layer underneath any agent. It captures decisions and drift from 200+ tools into one graph that Claude Tag, ChatGPT, Cursor, and Gemini all read and write.
• The decisive gap is temporal correctness: Claude Tag's launch describes no way to know when a fact stopped being true, while Sentra is bi-temporal (40% / 43% on KAIST's MEME benchmark vs a 3% / 1% field), so agents never restate a deprecated price or policy.
• The second gap is coverage: Claude Tag learns from Slack alone, while Sentra reads 200+ tools — email, CRM, code, and meetings — and most company knowledge never reaches Slack.
• Claude Tag executes; Sentra remembers, correctly and across the whole stack. The best setup runs both, and your memory stays portable across any model rather than locked to one vendor.

What We're Actually Comparing

Claude Tag and Sentra solve different problems, and the comparison only makes sense once you separate the two jobs. Claude Tag is an agent. You tag @Claude in a Slack channel, it breaks the task into stages, works through them with connected tools, and posts the result in the thread (anthropic.com). Sentra is the memory layer beneath any agent. It captures decisions, commitments, and facts into one graph that Claude Tag, ChatGPT, Cursor, or Gemini can read and write through an API or MCP.

Anthropic launched Claude Tag on June 23, 2026, in beta for Enterprise and Team customers, running on Opus 4.8 and replacing the older Claude in Slack app, which retires August 3, 2026 (anthropic.com). The orchestration is genuinely strong. The question for a buyer is not whether Claude Tag executes well, because it does. The question is what knowledge it draws on when it executes, and where that knowledge lives. Two gaps decide it in practice: how much of the company the memory actually covers, and whether it knows when a fact stopped being true. Vendor portability is a third, smaller consideration. That distinction between agent and memory layer runs through the comparisons that follow.

Quick Comparison

Claude Tag and Sentra answer different questions. Claude Tag runs tasks inside Slack and works through them autonomously. Sentra holds the memory that any agent, including Claude Tag, reads from and writes to. The table above tracks where each one leads and where the other fills the gap.

How We Evaluated Them

We judged both products on the four things that decide whether company memory holds up in production: memory breadth, temporal correctness, lock-in risk, and governance. Breadth determines what an agent can know when it answers a cross-functional question. Temporal correctness determines whether it restates a fact that stopped being true. Lock-in determines whether your memory survives a model switch. Governance determines whether a regulated buyer can deploy it at all.

One caveat on fairness. Claude Tag launched June 23, 2026, so every claim about its memory here comes from Anthropic's launch coverage, not assumed internals. Where launch documentation is silent, we say so rather than guess.

Memory Scope: One Source vs. the Whole Stack

Claude Tag learns your company from Slack, and nothing else. Its memory comes from the messages in channels it sits in, plus other channels an admin permits it to read (techcrunch.com). It does not access private channels, and launch coverage describes no ingestion from email, documents, code repositories, or CRMs (anthropic.com). For a team whose decisions, debates, and context genuinely live in Slack, that is a strong source. For everyone else, it is a partial one.

Sentra reads the whole stack. It connects to 200+ tools, including Gmail and Outlook, Google Drive, Notion, HubSpot, GitHub, Linear, and meeting transcripts from Fireflies and Granola, then resolves all of it into one shared graph. The breadth matters because most company knowledge never reaches Slack. A contract term might sit in email while a schema decision sits in a GitHub pull request.

The breadth decides what an agent can actually answer. Ask either system a cross-functional question, such as why a feature shipped late and which customer it affected, and the answer depends on what the system can see. Claude Tag can reconstruct the parts of that story discussed in Slack. It cannot reach the original commitment made over email or the code change that caused the slip, because those sources sit outside its memory.

Sentra also unifies identity across those sources, matching "Sarah Chen" in HubSpot to "@schen" in Slack and "S. Chen" in Gmail as one person. Without that resolution, an agent reading multiple tools treats one customer as three. Sentra gives an agent one memory to query for the full picture, rather than one channel's slice of it.

Temporal Correctness: Does It Know When a Fact Expired?

Most company memory treats every stored fact as currently true. A pricing change or a renegotiated contract term overwrites or sits alongside the old version, and the agent restates last quarter's number as if it still holds. Sentra solves this with bi-temporal awareness. Every fact carries when it became true and when it stopped being true. Old facts are invalidated, not deleted, so the graph keeps a complete history while only surfacing what is currently accurate.

When an agent asks Sentra about a customer's contract status, it gets the active term plus the provenance of what changed and when. Sentra also surfaces what has gone stale, what is at risk, and what has not been mentioned in two weeks, so drift becomes visible instead of silent.

The MEME benchmark from KAIST measures exactly this kind of temporal reasoning, and Sentra is the only system above 30% on both the Cascade and Absence categories, scoring 40% on Cascade and 43% on Absence. Field averages sit at 3% on Cascade and 1% on Absence. Sonnet 4.6 scores 5% on Cascade and 35% on Absence in the same test.

Claude Tag's launch coverage describes its memory as building "ever more context" over time, but no documentation details staleness handling, contradiction detection, or any mechanism for knowing when a fact stopped being true (anthropic.com). The architecture simply is not public yet. Until Anthropic publishes how Claude Tag retires outdated facts, you cannot assume it does, and a Slack-sourced memory with no documented temporal model will tend to treat the most recent mention as the current truth.

Model Lock-In: Tied to One Vendor vs. Portable

Claude Tag's memory lives inside Anthropic. It runs on Opus 4.8, and the context it builds from Slack stays bound to Claude. Nothing in the launch coverage describes that memory being readable by another model or agent. For a team standardizing on Claude Enterprise, that is a reasonable trade. The risk shows up later, when you want ChatGPT, Gemini, or a coding agent to draw on the same organizational knowledge and find it locked behind one vendor's identity.

Sentra holds the memory outside any single model and lets every agent read and write it. The graph connects over REST or MCP, and the named compatible tools include Claude, ChatGPT, Cursor, Perplexity, Codex, and Windsurf. The principle Sentra states plainly is that what you teach one agent, every agent remembers. A commitment captured while Claude Tag worked a Slack thread becomes available to Cursor during a code review and to ChatGPT when someone drafts a customer reply.

The vendor-risk case is practical, not ideological. Models change fast, and their pricing and availability shift with them. When your company knowledge sits inside the model, switching vendors means rebuilding context from scratch. When the knowledge sits in a shared graph, you swap the agent and keep the memory. Sentra treats the model as a consumer of memory rather than its owner, so you keep your company's context instead of renting it.

Commitment Tracking and Contradiction Detection

Sentra tracks every commitment from the moment someone speaks it, attaches the evidence behind it, and surfaces slippage before a deadline passes. In one example, Sentra tracked six commitments across four design partners. Four shipped, one slipped (SAML SSO for Acme), and one was dropped. That action memory layer also flags what has gone stale and what has not been mentioned in two weeks, so a missed promise surfaces as a risk instead of an after-the-fact surprise. Contradiction detection works the same way. When a new statement conflicts with an existing fact, Sentra catches the clash rather than storing both as equally true.

Claude Tag's launch coverage describes no equivalent capability. Anthropic documents @-tag delegation, shared channel identity, and persistent memory built from Slack messages, but nothing about tracking commitments over time or detecting contradictions across sources. That is a feature gap, not a knock on the agent's intent. Claude Tag executes tasks well. It just does not yet carry an action memory layer that watches a promise from the day it is made to the day it ships or slips.

Governance and Deployment

Claude Tag gives administrators tight control over what each Claude identity can touch. Admins define which tools, data sources, and channels a given Claude can access, set token spend limits at the organization and channel level, and review a full audit log of every action and who requested it. Memories stay scoped to admin-defined channels, so a sales Claude cannot seed memories into engineering, and Claude Tag never reads private Slack channels. For a Slack-centric team, that control surface is enough.

Regulated buyers usually need more than channel scoping. Claude Tag runs as a hosted Anthropic service, and its launch coverage names no compliance certifications, no isolated deployment, and no statement on whether your data trains future models. Those gaps matter when an auditor asks where company data lives and who can train on it.

Sentra answers those questions directly. It holds SOC 2 Type II and ISO 27001, deploys in the cloud, an isolated VPC, or fully air-gapped on-premises, and does not train models on your data. A bank or hospital that cannot send context to a third-party cloud can run the company graph inside its own boundary and still feed any agent through the API or MCP.

Claude Tag governs access well within Slack, while Sentra governs where the memory itself lives and who can use it.

Best For: Claude Tag

Claude Tag works at production scale, and Anthropic proves it. The company reports that 65% of its product team's code now comes from its internal version of Claude Tag, which also handles product metrics, support tickets, and bug root-cause analysis (anthropic.com). That is a strong signal for a tool in beta.

Pick Claude Tag if your team lives in Slack and you want an autonomous coworker you can use today. Tag @Claude in a channel, and it breaks the task into stages, works through connected tools, and posts results in the thread. Anyone on the team can pick up where someone else left off.

The fit is sharpest for organizations already on Claude Enterprise or Team, since Claude Tag ships to those plans first and runs on Opus 4.8. If most of your operating context already flows through Slack conversations, Claude Tag reads that context natively and acts on it without a separate setup step.

Best For: Sentra

Pick Sentra when your context lives in more than one place and your team runs more than one model. The buyer who needs Sentra most has commitments in HubSpot, decisions in GitHub, and customer calls in Fireflies, with threads scattered across Gmail and Slack. No Slack-only agent can answer a cross-functional question when the answer lives in four other tools.

Regulated industries also land here. SOC 2 Type II, ISO 27001, and air-gapped or VPC deployment let you run a shared company graph without sending data to a model vendor for training.

Multi-model shops gain the most. If your engineers use Cursor, your support team uses ChatGPT, and you also run Claude Tag, Sentra gives all of them one graph to read and write. What you teach one agent, every agent remembers.

Sentra does not replace Claude Tag. It feeds Claude Tag correct, cross-tool, bi-temporal memory over MCP, so the agent you already trust draws on more than Slack. Sentra makes whichever agent you run sharper.

Using Sentra and Claude Tag Together

Sentra and Claude Tag fit together because one supplies memory and the other executes against it. Claude Tag reads and writes Sentra over MCP, so its agent draws on a knowledge base that spans your whole stack instead of Slack channels alone. What Claude Tag cannot see in Slack, Sentra has already captured from email, GitHub, your CRM, and meeting transcripts.

Consider a task you delegate by tagging Claude in a channel. The work touches a renewal commitment a rep made over Gmail, an architecture decision logged in a GitHub pull request, and a customer call recorded in Fireflies. None of those live in Slack, so a Slack-only agent answers from a partial picture. Sentra resolves all three into one graph, attaches provenance, and marks the commitment as live or slipped. Claude Tag pulls that context through MCP and acts on facts that are current and correct.

When the renewal date changed last week, Sentra invalidated the old value rather than leaving two answers in circulation. Claude Tag then restates the date that is true today, not the one that expired. You keep Anthropic's agent and its Slack-native delegation, and Sentra removes the blind spots that come from learning a company one Slack message at a time.

FAQs

Does Claude Tag replace Sentra?

No. Claude Tag is an agent that orchestrates and executes tasks inside Slack, while Sentra is the memory layer that any agent reads and writes. Sentra gives Claude Tag a cross-tool, bi-temporal knowledge base it cannot build from Slack messages alone.

Can Sentra work with Claude Tag?

Yes. Sentra exposes its company graph over MCP, the same protocol Claude Tag uses to reach connected tools. Claude Tag executes the task and Sentra supplies correct, current facts from email, CRM, code, and meetings, so its answers reflect more than the channel it was tagged in.

What happens to Claude in Slack on August 3, 2026?

Anthropic retires the Claude in Slack app on that date and replaces it with Claude Tag, with a 30-day admin opt-in migration window (9to5mac.com). Sentra is unaffected by this change, since it runs as a model-agnostic memory layer rather than a Slack app.

Does Sentra store Slack messages?

Sentra ingests Slack as one of 200+ sources, then resolves the meaning at write time into a single graph alongside Gmail, Notion, GitHub, Linear, and HubSpot. It does not train models on your data, and it invalidates old facts rather than deleting them, so provenance stays intact.

Which is better for regulated industries?

Sentra carries SOC 2 Type II and ISO 27001 certifications and deploys in an isolated VPC or fully air-gapped on-premises, which matters when data residency is a legal requirement. Claude Tag offers admin-scoped channel permissions, audit logs, and token spend limits, and it does not read private Slack channels. Teams that need on-premises control will find Sentra the stronger fit, and the two can run together.