CTO Q&A · status as of 21 Oct 2026

How is BrightGears different from Claude in Slack?

Claude Tag is Anthropic's agent in Slack. BrightGears is something else: your own multi-agent software factory, installed on your server, tuned to your team and run together with you. Fourteen questions a CTO asks, answered with the real status of each item.

live on the production install today
partial works with limits
planned designed or decided, not built
gap not built

What is Claude Tag, and what happened to Claude Code in Slack?

Claude Tag gives every company the same hosted agent. BrightGears gives one team its own factory.

BrightGearslive

Your own multi-agent software factory: installed on your server, tuned to your team and run together with you.
Part software, part consulting. Our specialists set it up with you, watch how it works and keep improving it.
Few clients on purpose; each install is thought about.

Claude Tag

Anthropic's agent in Slack. One hosted agent, the same for every company, in Anthropic's cloud.
Legacy Claude Code in Slack is being retired: Pro and Max only, one person per session, GitHub only.
Retired on 5 Oct 2026 in workspaces paired with a Claude org (no data migrates); it still works for Pro/Max in unpaired workspaces.

Where does it run, and where do my code and conversations go?

Everything runs on your own server. Code, conversations and decisions stay there; nothing goes to a vendor's transcript store.

BrightGearslive

One install per client, on the client's server or cloud account; operator admin over a private network.
Persistent per-thread working folder on the client's machine.
Inside the client network, no proxy hop.
Regulated data stays on client infrastructure; compliance docs not yet written. No retention TTL policy yet.

Claude Tag

Anthropic cloud; ephemeral sandbox per thread, rebuilt on the next reply. Files kept only if pushed or posted.
Default-deny egress; internal services through federated access (customer gateway, cloud identity, OAuth).
Transcripts kept until deletion during beta, also after Slack deletes.
Not covered by BAA; unavailable with ZDR, CMEK or HIPAA config.

How is it billed?

The factory runs on the AI subscriptions your team already pays for. We have not modelled the cost difference yet, so we make no “10× cheaper” claim.

BrightGearslive

The client signs in its own Claude and Codex accounts from the web UI; gears rotate across them and wait out limits.
Switches automatically to the next subscription; pauses only when all are at their limit.
Fallback: API keys at API rates.
Risk: consumer plan terms forbid automated or bot access; Team/Enterprise terms are unclear.

Claude Tag

Team and Enterprise plans only.
Channel work is paid from an org usage balance at list price (the pricing page says API rates); seats do not cover channel work. A one-to-one DM uses the member's seat.
Team needs purchased credits; Enterprise is invoiced. Setup preselects a $2,500 monthly limit (options $500–$5,000, unlimited, or custom).
Per-token rates for this product are not published.

Am I locked to one model vendor?

No. We pick the best models and agent harnesses on the market and make them work together. One reviews the other's work.

BrightGearspartial

Claude runs the turns; Codex runs as a helper on its own pooled logins, useful for cross-provider review.
Codex as a full turn runner is planned. More providers as they appear.
When a better tool comes out, you get it.

Claude Tag

Anthropic models only (Opus, Sonnet).
Model switching and fast mode.

Is it one agent or a team?

A team. Each thread's gear runs helpers and multi-step workflows, and you see each agent's progress in one status line.

BrightGearslive

Leads, helpers and reviewers per thread: writers, critics, reviewers.
One small status line, always last in the thread: state, team, waits, next check.

Claude Tag

Sub-agent spawning is not documented.
Status in thread: a live checklist.

What stops an open task from being silently dropped?

A mandatory next check on every open thread. Closing a thread needs a stated reason.

BrightGearslive

Thread = card: title, why, now, next, owner, next checks.
Each open thread keeps at least one scheduled follow-up: when the gear looks again and why.
New messages enter the running turn at once.
Routines run at fixed intervals plus a health round; wall-clock schedules are planned.

Claude Tag

Also follows up: PR subscriptions, channel and thread follows that end after 7 quiet days, follow-ups until an approval lands.
A next check is not mandatory per task.
Natural-language routines, in UTC.

How does it learn the way my team works?

At setup we adapt the factory to your language, habits, review culture and autonomy. After that, gears propose habits and people confirm them.

BrightGearslive

Gears propose habits from how the team works; a person confirms; every gear in the install follows them on the next turn.
A person-only flag turns a habit into a locked auto-approve rule (today each change needs one admin command).
Quality signal: happiness score per member message, Pulse, thread metrics.
Lessons from closed threads, 7-day client review before upstream (partial).

Claude Tag

Editable memory notes per channel and workspace that anyone in the channel reads and corrects; custom instructions.
A skills repository where Claude proposes skill improvements as PRs that a person reviews.
Per-channel analytics.

How do I control what the agents may do?

You decide, down to the exact command a risky action needs approval for.

BrightGearslive

Risky actions wait for a person to approve the exact command, with a use count and expiry; first answer wins; via the web UI.
Locked rules: managed settings mean gears cannot widen their own permissions or hooks.
Roles: owner, approver, viewer.
Planned: an autonomy dial per team (from manual merge to autonomous merge) and Slack approval buttons.

Claude Tag

Auto mode with pre-approved rules for sandbox actions.
Review options for personal connectors and results; no exact-command gate.
Admin-scoped rules; workspace admins, per-channel control on Enterprise.

Do threads know about each other?

Partly, today. Each turn sees what other threads in the channel are doing; an install-wide digest is planned.

BrightGearspartial

Each turn gets its own thread's new events plus short lines on what other threads in the channel are doing.
Install-wide digests and thread search are planned.

Claude Tag

Isolates thread sandboxes but shares context: a channel-level session, hand-off of new messages to working sessions, past transcripts on request, following other channels.

What does it connect to: tools, repos, files?

Whatever your machine reaches. Managed connectors are not built yet; this is where Claude Tag is ahead.

BrightGearspartial

Connectors: whatever the client's machine reaches; managed connectors not built (gap).
Repos: any git host the machine reaches; GitHub App integration is on the backlog.
Files in Slack: members' attachments land in the thread folder (live).

Claude Tag

15+ managed MCP connectors (Jira, Notion, Datadog and more).
GitHub and GitLab out of the box, including GitLab self-managed.
Files in Slack: yes.

SSO, audit logs, compliance?

Not built. This is an open client ask, and Claude Tag is ahead here today.

BrightGearsgap

No SSO, audit export or SLA yet.
Sign-in over the private network today; Slack sign-in planned.
Open asks we have not answered: cost reporting, SSO, audit export, retention policy, off-site backup, compliance docs, subscription terms.

Claude Tag

SSO on Team and Enterprise; audit logs and SCIM on Enterprise; Activity page on all plans.
Slack Connect and guest-channel controls.

Which chat channels?

Slack today, Telegram with limits, Microsoft Teams researched.

BrightGearslive

Slack: every member message in an invited channel; silent on talk between people.
Telegram: yes, one chat = one thread; topic router designed (partial).
Microsoft Teams: researched, 1.5–2.5 weeks, needs a public HTTPS endpoint.

Claude Tag

Slack only.
Reads every channel message; it can reply without a mention; a mention only guarantees a reply.

What is the self-improvement story?

Simulated twins of your team and automatic A/B tests, under our own supervision. Twins are live in their first form; A/B tests are in build.

BrightGearspartial

Each team member has a simulated twin whose behaviour profile is distilled daily from their real messages. It can drive test threads without authority.
Happiness scoring of every message and reviewed lessons are live.
Planned: live A/B tests of preferences on new threads (keep the winners), closed-loop evals, cross-client learning.
One core for all clients; the operator rolls updates per install today; signed self-updates planned.

Claude Tag

A self-improving skills repo: Claude proposes skill improvements as PRs; a person reviews; every channel gets the change.
A/B tests: not documented.

Where is Claude Tag ahead today?

Eight places. We list them so the comparison stays honest.

Zero setup. Admin pairs the workspace; no server, no install, no operator.
Connectors. 15+ managed MCP connectors (Jira, Notion, Datadog and more). BrightGears has none built in yet.
Memory people can edit. Channel and workspace notes that anyone in the channel reads and corrects.
Routines in natural language. Our routines use fixed intervals; wall-clock schedules are planned.
Vendor polish and support. SSO on Team and Enterprise, audit logs and SCIM on Enterprise, an Activity page, model switching and fast mode, hosted artifact pages, Slack Connect and guest-channel controls. We have no SSO, audit export or SLA yet.
GitHub and GitLab out of the box (GitLab self-managed too). Gears' repo access is on the backlog.
A self-improving skills repo. Claude proposes skill improvements as PRs; a person reviews; every channel gets the change.
Reads every channel message. It can reply without a mention; a mention only guarantees a reply.

Full comparison

Every area, side by side.

BrightGears, Claude Tag (new) and the legacy Claude Code in Slack app. The last column is BrightGears' own status.

Area	BrightGears	Claude Tag (new)	Claude Code in Slack (legacy)	BrightGears status
Where it runs
Hosting	Client's server or cloud, one install per client; operator admin over a private network	Anthropic cloud; ephemeral sandbox per thread, rebuilt on next reply	Claude Code on the web sandbox	live
Workspace state	Persistent per-thread working folder on the client's machine	Files kept only if pushed or posted	Cloud session	live
Private network	Inside the client network, no proxy hop	Default-deny egress; internal services through federated access (customer gateway, cloud identity, OAuth)	Cloud only	live
Regulated data	Stays on client infrastructure (compliance docs not yet written)	Not covered by BAA; unavailable with ZDR, CMEK or HIPAA config	—	partial
Cost
Billing	Client's own subscriptions, rotated; API keys as fallback	Org balance at list price; $2,500/month preselected limit; DMs on the member's seat	User's Pro/Max subscription	live
Usage limits	Switches automatically to the next subscription; pauses only when all are at their limit	Spend limit	Subscription limit	live
Work model
Intake	Every member message in an invited channel; silent on talk between people	Reads every channel message; may reply without a mention; a mention guarantees a reply	Tag the app	live
Unit of work	Thread = card (title, why, now, next, owner, next checks)	Thread session	Session	live
Follow-ups	Mandatory next check per open thread; close needs a reason	Routines, PR subscriptions, channel and thread follows, follow-ups until an approval lands	—	live
Live steering	New messages enter the running turn at once	Reply in thread	Reply in thread	live
Status in thread	One small status line, always last: state, team, waits, next check	Live checklist	Checklist + web transcript	live
Agent team	Helpers and multi-step workflows per thread, shown in status	Not documented	No	live
Schedules	Routines (interval), health round, per-thread checks	Natural-language routines, UTC	—	partial
Context and learning
Cross-thread context	Each turn sees other threads' recent activity in the channel; install-wide digest planned	Isolated sandboxes; channel session, hand-off to working sessions, past transcripts, other channels	None	partial
Team knowledge	Preferences: proposed by gears, confirmed by people, applied to every gear	Channel and workspace memory notes	None	live
Quality signal	Happiness score per member message, Pulse, thread metrics	Per-channel analytics	—	live
Lessons	Lessons from closed threads, 7-day client review before upstream	Skills repo: Claude proposes skill PRs, a person reviews	—	partial
A/B tests	Preferences tested live on new threads; keep winners	Not documented	—	planned
Simulated users	Per-member twin, profile distilled daily from real messages, no authority	—	—	partial
Closed-loop evals	Sims drive test tasks; regressions blocked	—	—	planned
Cross-client learning	Reviewed lessons become core defaults for all installs	—	—	planned
Models and tools
Providers	Claude turns; Codex helpers on pooled logins; Codex turns planned	Anthropic only (Opus, Sonnet)	Anthropic only	partial
Connectors	Whatever the client's machine reaches; managed connectors not built	15+ MCP connectors	—	gap
Repos	Any git host the machine reaches; GitHub App integration on backlog	GitHub and GitLab (incl. self-managed)	GitHub only	partial
Files in Slack	Members' attachments land in the thread folder	Yes	—	live
Control and safety
Approvals	Exact command, use count, expiry, first answer wins; web UI	Auto mode + allow rules for sandbox actions; review options for personal connectors and results; no exact-command gate	—	live
Locked rules	Managed settings: gears cannot widen their own permissions or hooks	Admin-scoped rules	—	live
Roles	Owner, approver, viewer	Workspace admins; per-channel control on Enterprise	—	live
Autonomy level	Set per team: manual merge to autonomous merge	Per-scope rules	—	planned
Sign-in	Private network today; Slack sign-in planned	Slack + Claude org	Claude account	partial
SSO, audit export	Not built	SSO on Team and Enterprise; audit logs and SCIM on Enterprise; Activity page on all plans	—	gap
Data retention	Client controls the machine; nothing stored at a vendor; no TTL policy yet	Transcripts kept until deletion during beta, also after Slack deletes	—	partial
Channels
Slack	Yes	Yes	Yes	live
Telegram	Yes (one chat = one thread; topic router designed)	No	No	partial
Microsoft Teams	Researched: 1.5–2.5 weeks, needs a public HTTPS endpoint	No	No	researched

Sources for Claude Tag and the legacy app: claude.com/docs/claude-tag (overview, how-it-works, memory, proactivity, customize, set-spend-limit, security-and-data, data-lifecycle, healthcare, migrate-from-earlier), code.claude.com/docs/en/slack, claude.com/pricing; read 21 Oct 2026. BrightGears status: core main ef1eaa1 and the feature inventory.

See the factory run on your tickets.

Few clients on purpose. Every install is set up with you and improved every week.

Book a fit call