Agents & Actions

Agents that answer — and then act.

Configure a custom agent, wire it to the tools it needs, and let it take real, governed actions on your behalf.

Renewals Agentrunning on your chosen model
BUILDRUN
Agent
Knowledge search
Web search
Send emaildone
“Email the Q3 renewal summary to the Acme owner.”
acted as: you
Fires once Acts as you Full audit log
Why it matters

Measured in outcomes, not output.

A chatbot returns text; an Augmas agent finishes the task and leaves a result you can point to.

Output
“Here’s what you could do next…”nothing happens
the gap
Outcome
Ticket filed#4821
Event createdThu 3:00 PM
Email sentto finance@
Update posted#ops-room
Build an agent

Give it instructions, a scope, and the tools to act.

Every agent is a few deliberate choices — instructions, starter messages, a document scope, a model, and the tools it may use.

New agentModel Anthropic
Instructions
You are the Renewals Agent. Draft in our voice; cite the record; never invent numbers.
Starter messages ×3
Summarize this accountDraft a renewal emailWhat’s the risk here?
Document-set scope
Sales workspace4 sourcesscoped
Linked tools 5
Knowledge searchSend emailManage calendarsFile ticketsWeb search
Renewals Agent
Anthropic · 5 tools · scoped
Save agent
The tool palette

A palette of actions you attach.

Snap on the built-in tools every agent ships with, plus connected actions like email, calendars, tickets, and updates.

Built-in — ships with every agent
Knowledge search
Web search
Code execution
Image generation
Connected actions
Send emailvia connector
Manage calendarsvia connector
File ticketsvia connector
Post updatesvia connector
Renewals Agent
5 tools attached

attach = the agent may use this tool

Custom tools

Connect any API. Zero code.

Point Augmas at an OpenAPI spec and your REST endpoint becomes an agent tool — no engineering change.

Your API
POST/v1/refunds
GET/v1/orders/{id}
OpenAPI spec
drop spec
createRefund
getOrder
Agent tool
Refunds tool2 operations
no code changesCustom MCP tools supported too
The run loop

Reason, act, observe — one clean loop.

The agent thinks, picks a tool, calls it across every connected source, and observes the result — grounding each answer in cited evidence.

GUARDED
reasoning loop
01Reason
02Pick tool
03Call
04Observe
1
+46

queries every connected source at once

2
Side-effectguarded execution fires once
3
Answer · cited

Acme renewal is medium risk — one blocker on the critical path, and the sponsor changed in June.

Jira · ACME-482Slack · #renewals
no evidence → no answer
Guarded loop

Every side-effect fires exactly once.

Sending an email or creating an event happens a single time — retries and network blips can never duplicate it.

Send emailFired · once
Retry attempt · network blipalready done · will not resend
Ledger · exactly one entry · Send email · 14:02
Acts as you

Your agent acts as you — never a shared login.

Each person connects their own Google, Atlassian, or GitHub account inline, and the agent acts with exactly that user’s permissions.

Your accounts
each user connects their own
GGoogleconnected as alex@acme.comyour permissions
AAtlassianconnect
GHGitHubconnect
How the action runs
same agent, each person’s identity
ALAlex → sends as Alexalex@acme.com
SASam → sends as Samsam@acme.com
shared service account
Personal assistants

A personal assistant for everyone.

Every user gets their own assistant agent out of the box — and can spin up as many personal agents as they need.

Your personal assistants

One for everyone. As many as you want.

Fits your stack

Reachable in Slack, on the model you choose.

Talk to your agent right in Slack, and set each agent’s model per agent — from major providers to a local option.

In Slack
#ops-room
DM
Dana M.

@Renewals Agent post the deploy status

Renewals Agent

Deploy #241 is green across all checks. Summary posted to the release note.

Update posted · via Slack
On your model
Anthropic
Anthropic
OpenAI
Azure OpenAI
Gemini
Groq
Ollamalocal

Swap the model — keep the instructions, scope & tools.

Governance

Every action, on the record.

Scoped permissions and per-user identity are backed by a full audit log and enterprise controls, deployed inside your own environment.

Action audit log
14:02alexSend email · Gmailfired once
14:07samFile ticket#4821
14:11alexCreate eventThu 3:00 PM
14:19danaPost update · Slack#ops-room
RBAC
SSO / SAML
SCIM 2.0
AES-256
Dedicated deployment in your environment
Action controls: scoped permissions · per-user identity · fire-once · full audit

Put an agent to work.

Configure your first agent, wire it to your tools, and let it take governed action from day one.

Agent · liveall tools docked
Send emailfired ✓