One shared AI agent for your whole team, not an AI seat per person.
Runbear runs Claude and GPT as shared AI agents in Slack and Microsoft Teams. One agent serves the whole team, and tools connected with per-user auth act with each person's own account.
A shared AI agent is one AI teammate that your whole team uses in Slack or Microsoft Teams, instead of a private chatbot in each person's browser tab. Runbear runs Claude or GPT as that named teammate: one agent serves the team, and tools connected with per-user auth act with the requester's own account. Across 40+ customer workspaces, 60%+ of seats are still active weekly at 30 days. That's adoption that compounds, not seat utilization.
The permission layer Anthropic doesn't ship.
Claude Enterprise gives you per-seat licenses. Runbear gives your shared Claude agents per-user authentication, plus audit logs and SAML SSO on the Business and Enterprise plans and SCIM provisioning on Enterprise — the structural feature your security team will ask for before signing.
Used by Matillion and other Anthropic-customer mid-market SaaS teams.
Shared agent identity
One named teammate the team @-mentions in Slack and Teams — not 15 private chat tabs that never share state. Usage-priced, so adoption is what compounds, not seat utilization.
Per-user authentication
Shared identity, per-user permissions. Tools connected with per-user auth, such as Salesforce or GitHub, act with the requester's own account instead of one shared service account.
Audit log
Every configuration change, with who made it, from where, and when, exportable as CSV or JSON for your SIEM. The agent's Activity tab shows its runs.
SSO + SCIM
SAML SSO on Business with identity providers such as Okta or Azure AD. Enterprise adds SCIM provisioning, so off-boarding flows through the same identity layer as the rest of your stack.
Hours reclaimed, FTE recovered, dollars saved — broken down per named agent on a dashboard your CFO can defend.
Catch at-risk accounts 6-8 weeks before the renewal call.
Runbear lives in your CS channels. It surfaces churn signals before they hit the renewal call, drafts QBRs from your product-usage data, and gives non-technical CSMs leverage on 40+ accounts each.
Teams already running Runbear
- Matillion
- Fireflies.ai
- BCG
- Minted
- Mercor
- Tabs
- TaxDome
- Peak Ace
- Storykit
- SYPartners
Personal Claude is brilliant for one person. It doesn't show up in the channel where your team is actually working.
Every workflow that matters to a team happens in a shared channel: CS questions in #cs-questions, deal context in deal channels, on-call paging in #incidents, HR questions in #ask-people. A personal Claude tab lives outside those channels. People have to copy the question out of Slack, paste it into a browser, paste the answer back. The friction is the whole story: in the moments your team would have used Claude, they don't, because the assistant isn't where the work is. Runbear puts the assistant in the channel, named and addressable, with the team's context built in.
One agent serves the whole team in the channel they already use, instead of 15 private chat tabs that never share state.
When each teammate runs their own personal Claude, the team's institutional knowledge stays scattered across 15 browser tabs. Each tab forgets what the others learned. With Runbear, the agent has a name (RunbearCS, RunbearOps, whatever your team chooses) and a single conversation log the team can search. New hires learn the team's preferences by reading what the agent has answered before, in the same Slack channel the team already uses for everything else.
The agent is shared. For tools connected with per-user auth, the permissions are per-user: the agent acts with the requester's own account, not one shared service account.
This is the security property that lets a shared agent be safe. Each connected tool is authorized either Per-User or Shared. With Per-User, each member links their own account, so a finance analyst asking the agent for a deal record in Salesforce only gets what they could see in Salesforce directly, and a teammate without that access gets a scoped answer. Shared tools use one account for everyone, so grant them deliberately. Knowledge sources such as Notion or Google Drive sync through the account of the member who connected them, so connect them from an account whose access fits the channel. The shared identity is the agent's; per-user tools keep the team's existing permission model.
Three loops compound the team's institutional knowledge inside the agent: synced knowledge sources, in-channel corrections, and named agents that own a topic.
The agent answers from connected knowledge sources that Runbear keeps synced, and tools such as a ticketing system are read live on each request. When the team corrects an answer in the channel, the agent's memory can carry that correction into later answers. Named agents own a topic: RunbearHR knows the people policies; RunbearOps knows the runbooks; they don't bleed into each other. Six months in, the agent answers questions the team has answered before, the way the team answers them, with citations to the documents the team actually maintains.
Runbear charges by usage, not per seat. Because adoption is what compounds, not seat utilization.
Per-seat AI pricing is an adoption tax: you pay $30 per user whether or not they log in, and you have to drive every user to log in every month to make the math work. Most enterprise deployments do not hit that bar. Runbear charges by usage, so the team's shared agent serves everyone and what compounds is adoption: questions answered, hours reclaimed, FTE-equivalents recovered. Per-agent ROI views show hours reclaimed, FTE recovered, and dollars saved on each named agent, on a dashboard the CFO can defend.
There is no engineering ticket. A team lead describes the agent in plain English and connects the team's tools.
The typical rollout: a workspace admin installs Runbear from the Slack Marketplace (one install covers the whole workspace), a team lead describes the agent in plain English and connects the tools and knowledge sources it should use (Notion, Confluence, Google Drive, Salesforce, HubSpot, Zendesk, GitHub, and more), and the agent answers in the channels it is invited to from day one. Most teams have their first answer in production in about 10 minutes. The first 30 days is the adoption curve, not the build curve.
One shared agent. Many team shapes.
Across 40+ customer workspaces, Runbear powers the shared AI teammate inside teams as varied as 12-person CS shops and 200-person engineering orgs. The same shared-agent model serves HR teams answering benefits questions, on-call engineers correlating alerts, and AEs prepping deal briefs. The team's adoption curve runs steeper than the seat curve: 60%+ of seats stay active weekly at 30 days, because the agent is in the channel the team already lives in, not in a browser tab no one opens.
What we watch when teams roll out AI
We watch teams try personal Claude first. A few power users hit it hard. The rest of the team never opens it. Three months later the seat-utilization chart looks like the chart we expected from any per-seat AI license: a long tail of dead seats with a small group carrying the math.
Then the same teams move to Runbear. The agent shows up in #cs-questions, in #ask-people, in the deal channels people already live in. Adoption flips. The team that wouldn't open a browser tab for the AI assistant will answer in the same Slack channel they already use for everything else.
Two years of data across 40+ workspaces says the same thing. Teams that adopt Runbear inside their team's existing channels keep 60%+ of seats active weekly at 30 days. The pattern is the assistant going to the team, not the team going to the assistant.
When the agent doesn't know
When the agent doesn't know, it says so. Across our customer workspaces, about 15% of knowledge questions hit a "route to a human" path within 30 seconds. Another small slice return the wrong answer and get a thumbs-down from the channel. Both go into the correction loop: the agent reads the team's response, asks a follow-up if needed, and updates its handling for the next time the same shape of question shows up. The agent doesn't get smart on its own. The team's corrections make it smart.
Frequently asked questions
- SOC 2 Type IIin production
- Per-user authon shared agents
- Cancel anytimeno annual lock-in
- HIPAA Compliance (BAA on request)Eligible Enterprise customers
Where to read more next.
Runbear for project-based teams
Kick off client projects with contractors and experts who ramp up instantly. The team's context is in Slack from day one.
See the project-teams solutionRunbear for agencies
Every client team gets a named agent with the context the engagement needs. Onboarding new collaborators stops being a tax.
See the agency solutionRunbear for SaaS teams
Customer success, support, and engineering share one teammate in Slack. Adoption stays above 60% weekly at 30 days.
See the SaaS solution
Bring a workflow. Runbear ships the agent into your team's Slack by end of call.
Most teams have a first working agent in about 10 minutes. No engineering required, no procurement gate to start.
Want a walkthrough first? Book 20 minutes with our CEO