Product · AI agents

AI agents that act, not chat.

Grid's agents update deals, file tasks, send messages, and log activity. Inside your tenant. Through the same access layer your team uses.

Run a live agentSee the safety model

See an agent run.

Real tenant. Real tools. Read-only — your data stays untouched.

grid agent — stale-deals

prompt

Find deals stuck in negotiation for >7 days. Draft a follow-up task for each owner.

agent trace

cost$0.04steps0/8✓tenant-scoped
Open the full AI sandbox

The plan beneath the trace.

Every run breaks into nested steps with real Grid tools. Hover a subtask to read the reasoning — the prompt that picked the tool, the args, the result.

Plan view — illustrative. Real plans surface in the agent run audit on every run.

Agents that can't see what you can't see.

Every Grid AI tool runs through the same access layer your dashboard uses. No back-doors. No "admin mode" for the model. Same row permissions. Same tenant scope. Same audit.

Scope

Tenant-scoped at the database.

Every tool invocation is bound to your tenant before the query runs. The model can't reach another tenant's data — not by mistake, not by prompt injection, not by clever role-play. Enforced at the database boundary, not in the prompt. We have a linter that fails the build when an engineer forgets.

Per-record

Per-record, not per-tenant.

If your role can't update that deal — because it's owned by another team, or because you only have read scope — the agent can't either. Roles propagate from your seat into every tool call. Audit shows who asked, what ran, and what the model saw.

Forty-two tools. One agent loop.

Read your data. Score it. Update it. Draft from voice. Approve and assign. Research the web. Spawn a child agent. Remember the outcome.

  • Read

    Read what you can read.

    • crmContactsFind
    • crmContactsGet
    • crmCompaniesFind
    • crmCompaniesGet
    • crmDealsFind
    • crmDealsGet
    • crmNotesFind
    • tasksFind
    • documentsFind
    • documentsGet
    • foldersFind
    • foldersGet

    12 tools

  • Score & enrich

    Score it. Spot the risk.

    • crmContactsScore
    • crmDealsAnalyze
    • crmCompaniesEnrich

    3 tools

  • Write

    Write it back.

    • crmNotesCreate
    • crmNotesUpdate
    • tasksCreate
    • tasksUpdate
    • tasksComplete
    • tasksDelete
    • tasksAssign
    • crmContactsUpdate
    • crmCompaniesUpdate
    • crmDealsUpdate
    • crmActivitiesCreate

    11 tools

  • Automation

    Run it on a schedule. Or after a meeting.

    • tasksDraftFromTranscript
    • tasksSetCategory
    • tasksSetTags
    • tasksSetRecurring
    • tasksBulkUpdate
    • tasksBulkAssign
    • tasksRequestApproval
    • tasksApproveCompletion

    8 tools

  • Research & memory

    Research the web. Remember what worked.

    • webFetch
    • webSearch
    • linkedinLookup
    • memoryRemember
    • agentSpawn

    5 tools

Costs you can cap. Runs you can replay.

Eight steps per run. Two dollars per run by default. Raise it per tenant when you trust the agent — lower it when you don't.

  • Eight-step ceiling. The loop can't sprawl.
  • Two-dollar cap per run. Per-tenant override via `ai-tool-policies`. Per-tool budgets too.
  • Three audit lenses. Run → tool invocation → model call. Every step joined.
  • Replay any run. The prompt. Every tool call. Every result. Down to the model bill.
  • Monthly cost rollup, per tenant, per user, per tool.

Spawn sub-agents. Remember what worked.

An agent can spawn a child for a focused sub-task. Both inherit the parent's access scope. Both write to the same audit trail. Nest up to three deep.

  • `agentSpawn` opens a child run with narrower scope.
  • Depth-3 cap. The child can't spawn a grandchild that spawns again.
  • `memoryRemember` persists a learning into tenant memory, scoped to the tenant.
  • A consolidation job runs nightly to dedupe and rank what's stored.
  • Future runs read the memory before planning — so the agent gets better at your data.

Compared to AI workspaces.

Notion AI summarizes documents. Glean finds them. Grid acts on them.

Notion AI
Summarizes documents
Searches your data
Limited
Updates a CRM record
Files a task with owner + due date
Limited
Drafts an email, books the follow-up
Inherits user permissions
Workspace
Hard cost cap per run
Credits
Three-lens audit per tool call
Glean
Summarizes documents
Searches your data
Updates a CRM record
Files a task with owner + due date
Drafts an email, books the follow-up
Inherits user permissions
Workspace
Hard cost cap per run
Enterprise
Three-lens audit per tool call
Limited
Grid AI
Summarizes documents
Searches your data
Updates a CRM record
Files a task with owner + due date
Drafts an email, books the follow-up
Inherits user permissions
Per-record
Hard cost cap per run
$2 default
Three-lens audit per tool call
Three-lens

Pairs well with

Ready to put agents on the work?

Talk to the teamStart free

no credit card · cancel anytime · 14-day trial · cost cap on by default