v2.8Undo window is now 72 hours

Agents that run the work between your tools.

Weftline gives operations teams a crew of AI agents that read from the systems you already use, plan the busywork, act inside the limits you set and report back with a receipt for every step.

  • Approval on anything that moves money
  • One-click undo on every write
  • US, EU or APAC hosting
Run 4,182 · Support triageLive
  • 14 tickets read
  • 11 replies drafted
  • 3 refunds waiting for you
Open the live run

01 How it works

From connected tools to finished work, in four moves.

No flowcharts to draw. You describe the outcome; Weftline turns it into a plan, runs it inside your limits and shows its work.

  1. 01 Connect

    Connect the stack

    Link your helpdesk, ledger, CRM and team chat with scoped access. Everything starts read-only, and most teams connect five tools in under 20 minutes.

    60+ connectors · OAuth or API key · read-only by default

  2. 02 Set goals

    Set the goal and the guardrails

    Describe the outcome in plain words, then set the limits: amounts, accounts, working hours and who signs off.

    goal: Refunds under $250 resolved within 4 h

  3. 03 Plan & act

    Agents plan, then act

    Each agent writes a plan you can read before anything happens, then works through it and pauses wherever your rules ask for a human.

    Plan first · approvals inline · every write reversible

  4. 04 Report back

    A receipt for every run

    Runs end with a receipt: what changed, where and why, with one-click undo. A daily digest lands in your team chat or inbox.

    Receipts per action · digest at 9:00 local time

02 Live run

Watch an agent work. Then decide.

A simulated run, streamed step by step. Approve what it flags, or undo the whole run with one click. Nothing on this page touches a real system.

Ready
run 4,183goal · clear refund queue under policy R-2elapsed 00:00
    Simulated run · nothing touches a real system

    03 Use cases

    Built for the teams that keep the company running.

    Start with the one queue that eats your week. Weftline ships with playbooks for the three most common ones.

    A support specialist in a headset smiles while working through tickets on a laptop
    Support desk

    Support operations

    Triage, refunds and escalations, handled inside your policy.

    A
    62%of tickets resolved without a handoff
    −41%first response time
    • Refund and credit triage
    • Replies in the customer’s language
    • Tier 2 handoff notes
    A hand with a stylus checks printed charts against a tablet next to a calculator
    Month-end close

    Finance close

    Reconciliations and journal drafts, ready for review on day one.

    B
    5 daysto close, down from 9
    1,284payouts matched per run
    • Payout and bank reconciliation
    • Draft journal entries
    • Variance commentary
    Two colleagues review a printed pipeline report beside an open laptop
    Pipeline review

    Sales operations

    A CRM you can trust on Monday morning.

    C
    0stale deals at forecast time
    37duplicates merged a week
    • Dedupe and enrichment
    • Lead routing in seconds
    • Pipeline hygiene nudges
    Also on Weftline
    • IT access reviews
    • Vendor onboarding
    • Payroll checks
    • Inventory alerts
    • Renewal prep

    Figures are illustrative targets for this concept, not customer results.

    04 Integrations

    Plugs into the stack you already pay for.

    60+ connectors across the categories ops teams live in, plus webhooks, SQL and a typed REST API for everything else. Hover a tile to see it light up in the hub.

    • Helpdeskread · write
    • CRMread · write
    • Ledgerread · draft
    • Billingread · write
    • Paymentsread · refund
    • Bank feedsread
    • Data warehouseread · query
    • Spreadsheetsread · write
    • Team chatpost · approve
    • Emailread · send
    • Calendarread · book
    • Docsread · write
    • File storageread · write
    • Issue trackerread · write
    • HR systemread
    • Identityread · revoke
    • Analyticsread
    • Status pageread · post
    • E-signaturesend · track
    • Procurementread · draft
    • Webhooksin · out
    • SQLquery
    • REST APItyped SDK
    • SFTPdrop · pick up
    60+connectors
    2-waysync on writes
    1 tokenscoped per agent
    0rate limits broken

    Icons show generic tool categories. No third-party products are implied.

    05 Security & governance

    Every action on the record. Every permission on purpose.

    Agents get the least access they need, ask before anything risky and leave a trail your auditors can actually read.

    Audit log

    Append-only · exportable
    • 10:42:31support.agentrefund.issueORD-48211approved · you
    • 10:42:30support.agentreply.sendT-4821allowed · R-2
    • 10:41:58finance.agentjournal.draftJE-2291awaiting
    • 10:41:12salesops.agentaccount.mergeACC-1180reverted · you
    • 10:40:47access.agentseat.flagadmin-07read only

    Every write stores its before-state, so any row can be undone and every decision traced to a person or a policy.

    Permissions per agent

    Tap to cycle

    A dim server hall lined with glass-fronted racks
    EU · Frankfurt + Dublin

    Data residency

    Sign-off rules

    • Refunds over $250team lead
    • Any journal entrycontroller
    • Deleting recordsalways asks
    • Writes after 8 pmpaused
    • Bulk edits over 500 rowsasks first

    No training on your data

    Prompts, records and outputs are never used to train shared models.

    SSO, SCIM and roles

    SAML or OIDC sign-in, automatic provisioning and roles for builders, approvers and viewers.

    Encrypted throughout

    TLS 1.3 in transit, AES-256 at rest and customer-managed keys on Enterprise.

    Retention you set

    Keep logs from 30 days to 7 years, and purge a workspace on request.

    Security and compliance features are described as a design concept. Weftline is not a real product and holds no certifications.

    06 Pricing

    Pay for seats and the work agents actually do.

    Seats are the people who set goals and approve; viewers are free. An action is one write an agent makes in another system. Reading and drafting never count.

    Estimate your month

    14,000
    Seats
    12
    Billing
    Best fitTeam$522per month, estimated
    Best fit

    Team

    For one team’s first workflows.

    $39per seat
    per month

    5,000 actions included, then $0.006 each

    Your estimate$0
    • Up to 25 seats
    • 20 connectors
    • Approvals by amount and type
    • 24-hour undo window
    • 30-day audit log
    Start a pilot
    Best fit

    Scale

    For ops orgs running agents every day.

    $89per seat
    per month

    50,000 actions included, then $0.004 each

    Your estimate$0
    • Unlimited connectors
    • Multi-step approvals and roles
    • 72-hour undo window
    • 1-year audit log with export
    • SSO and SCIM
    Start a pilot
    Best fit

    Enterprise

    For regulated and global teams.

    Custom

    Committed volume, from 50 seats

    Your estimateLet’s talk
    • US, EU or APAC data residency
    • Private networking and own keys
    • Retention up to 7 years
    • Dedicated solutions engineer
    • Quarterly governance review
    Talk to us

    Prices are illustrative for this concept and shown in USD before tax.

    Questions, answered.

    Still curious? Bring them to the demo.

    What counts as an action?

    One write in another system: a reply sent, a refund issued, a record merged, a journal posted. Reading, planning and drafting are free, and undoing an action never counts.

    Can agents do anything without approval?

    Only what you allow. Each agent has scopes for read, draft, act and money, each set to off, ask or allow, plus thresholds such as amounts, record counts or hours.

    How does undo work?

    Every write is logged with its before-state. Inside the undo window (24 hours on Team, 72 on Scale, custom on Enterprise) one click reverts a single step or a whole run.

    Where is our data stored?

    In the region you choose. Prompts, records, logs and embeddings stay in US, EU or APAC hosting and are not sent elsewhere for processing.

    Do you train models on our data?

    No. Your data is used only to run your agents and is never used to train shared models.

    How long does setup take?

    Most teams run a first supervised job within a week: connect tools on day one, write goals and guardrails on day two, then watch supervised runs until you trust them.

    07 Changelog

    Shipping every other Tuesday.

    Small, steady releases. Here is what changed recently.

    1. v2.8Nov 10, 2026

      72-hour undo window on Scale New

      Revert any write for three days instead of one. Bulk approve and bulk undo also arrive in the inbox.

    2. v2.7Oct 27, 2026

      Plan diffs Improved

      When an agent replans mid-run, you now see exactly which steps changed and why before it continues.

    3. v2.6Oct 13, 2026

      APAC data residency New

      Workspaces can now be hosted in the APAC region, with logs and embeddings kept in-region.

    4. v2.5Sep 29, 2026

      Cost guardrails New

      Cap actions per agent per day and get a heads-up in chat at 80% of any budget.

    5. v2.4Sep 15, 2026

      3× faster ledger matching Fixed

      Payout files over 50,000 rows now match in under two minutes instead of timing out.

    08 Book a demo

    Bring one messy workflow. Leave with a running agent.

    A 30-minute working session with a solutions engineer. We connect a sandbox copy of your tools and run your workflow live, start to finish.

    A small team around a table closes a working session with a handshake
    Working sessionYour team, one engineer and a sandbox. You leave with a goal, guardrails and a first supervised run.
    1. 01Pick the workflow
    2. 02Live session, 30 min
    3. 03Two-week supervised pilot
    Best time

    Concept form: nothing is sent or stored.