Your processes shouldn’t depend on reps remembering.
Renewals, handoffs, forecast hygiene. Every fix you’ve been sold starts with somebody else changing their habits. Opine works from the calls, emails and threads that already happen, so the record catches up on its own and you build from there.
4.8/5Trusted by high-performing revenue teams
Every hygiene push dies the same way.
“Our Salesforce hygiene, candidly, sucks. If they don’t do it, it doesn’t get done, and sales team does the minimum every time because they’re managing so many deals.”
A RevOps leader, in a first call
A renewals motion nobody follows. A handoff that drops half the evaluation. Qualification that happens once, at kickoff, and never again. You’re the one expected to fix these, and every fix you’ve been sold starts with somebody else changing their habits.
They won’t. That’s not cynicism, it’s ten years of evidence. Meanwhile the information those processes need already exists, sitting in the calls, the email threads, the Slack channels and the meetings.
So automate the reading instead of the typing. That’s a context problem before it’s a workflow one.
Pick the fields. Pick the sources. Walk away.
Everything you build later reads from the deal record, so start by fixing that. Each field gets its own sources, its own cadence, and its own rule about whether a value writes itself or waits for you. This is the whole setup, not a friendlier picture of it.
| Field | Reads from | Cadence | On a match |
|---|---|---|---|
| Competitors | Calls, email | Daily 02:00 | Auto-apply |
| Next step | Calls, Slack | Daily 02:00 | Auto-apply |
| Technical requirements | Calls, shared docs | Daily 02:00 | Auto-apply |
| Participant roles | Calendar, email | Daily 02:00 | Auto-apply |
| Economic buyer | Calls, Slack | Mondays 02:00 | Review |
| Decision criteria | Calls | Mondays 02:00 | Review |
| Close date | Not read | Not scheduled | Owner only |
“What does this cost to run across 800 deals?”
You pay for movement, not record count.
A run passes over any deal with nothing new since the last one, so the bill tracks what happened in your pipeline rather than how many rows are in it.
“How do I know a filled value is right?”
Every value shows its source.
Each fill cites the call or the thread it came from, so a review is a check against evidence instead of a judgment call about a model.
“What’s the blast radius if it gets one wrong?”
Your permissions, not ours.
Writes go through the acting user’s own CRM credentials, and a value that didn’t change never writes at all.
The scheduled part is the reading and the filling, which is the part nobody wants to do.
Opine → CRM instant · CRM → Opine ~10 min
You decide what it’s allowed to touch.
Nobody’s asking you to hand a model your system of record on day one. Every capability is set on its own. Flip the switches below and watch what changes.
Read-only work runs on its own. Anything that writes outside Opine waits for a person, until you decide it’s earned otherwise.
Every capability is set on its own
An agent can write deal notes freely while its CRM updates wait for a person. What we recommend and demo: read-only work automatic, writes gated, and gates relaxed as an agent earns them.
Auto-apply is opt in, field by field
No field writes itself just because another one does. You pick the ones that have earned it, one at a time, and the rest queue for review with the reasoning attached.
Writes carry your own credentials
CRM writes go through the acting user’s authorization, so the field-level security you already built applies unchanged. Opine doesn’t need a way around it.
Queries never write
Reporting runs as Opine SQL against your own data, read-only and scoped to whoever asked. Ask the same question twice and you get the same answer.
A forecast is only as true as the record under it.
The record rolls up into the forecast. The forecast rolls up into the number you defend in the board meeting. Three teams treat those as three problems. They’re one.
- The recordNorthwind Logistics · opportunity
Stage: Commit
Last touched by a rep 24 days ago. Champion field still names someone who left.
- The forecastQ1 commit roll-up
$4.10M
Every deal counted at the category its owner set.
- The reportWin/loss, quarter to date
Lost to pricing
Read from the one picklist a rep filled in at close.
- Your numberBook 30 minutes
See what yours looks like with the evidence attached.
Thirty minutes on your own pipeline shape: which fields you’d fill first, and what the roll-up reads once every deal carries its own.
Nobody made a mistake in the forecast or the report. The number at the top is wrong because the record at the bottom was a guess. Fix the guess and the other two correct themselves, which is why this is a context problem rather than three cleanup projects.
Your next automation is a sentence, not a project.
Hygiene is the first process you automate here, not the last. Four surfaces sit on the same context, and reaching them doesn’t take a support ticket, a services engagement, or a quarter of somebody else’s roadmap.
Agents
Describe the job. Opine writes the agent.
Say what you want done in plain English and an assistant drafts the instructions and picks the triggers. Then you read the run log: what set it off, what it touched, what it held for you, and what it cost.
Runs are sandboxed and survive a deploy mid-flight, so a Tuesday release doesn’t take your automation down with it.
Read 14 calls, 62 emails and 3 Slack threads on the deal.
- Filled Next stepSecurity review scheduled Mar 18 · from the Mar 12 call
- Flagged champion departureNo decision-maker contact in 24 days · thread, Feb 18
- Close date → Apr 30Held for your approval · CRM write gate is on
Break it on your data first
Run an agent against a dataset and watch where its judgment goes wrong before your pipeline finds out. Costs are prepaid per run, so there’s no overage bill waiting at the end of the month.
Creditsprepaid · per-run cost · auto-refresh
Your playbook, made executable
The discovery framework, the battlecards, the way your team runs a POV. Written once, and any agent can opt into it. Open SKILL.md format, so what you build here leaves the same way it came in.
FormatSKILL.md · zip in, zip out · opt-in per agent
Works in both directions
Query your pipeline from inside Claude or Cursor over OAuth, scoped to what that person is allowed to see. Opine is a client too, so the agents you build can call the servers your org already connects.
AuthOAuth · no static tokens · authored agents save disabled
Reporting runs as Opine SQL, read-only and scoped to whoever asked, and it’s the same engine behind the dashboards. Teams who would rather skip the UI entirely export against the schema catalog and join Opine with their CRM in the warehouse. Read the Agents docs
Build your first agent with us on the call
One of your processes, start to finish: what sets it off, what it is allowed to write, and what the evaluation looks like before it goes anywhere near production.
What you ship to everyone else.
Most people will never build an agent, and they shouldn’t have to. One or two of you configure the autonomy. Everybody else just gets the output.
Security review scheduled Mar 18
Follow up with Dana
Nothing new to type. The record caught up from work they had already done.
Northwind Logistics: champion has gone quiet
No contact with a decision maker in 24 days.
The alerts you configured, landing where they already read.
$3.26M
Down $840,000 after three deals lost their evidence.
The dashboard you built is the one they open.
- Pipeline hygieneLive
- POC handoffLive
- Renewal riskIn eval
A portfolio, not a campaign you run every quarter.
Adopted in full, integrated in a day.
Opine is the AI-ready context layer for revenue operations: it reads the calls, threads and documents your org already produces and keeps one current, evidence-backed record per deal.
The questions RevOps asks us first.
What it depends on, what it’s allowed to touch, and what it costs to run across the whole book.
Have one we didn’t cover? Bring it to the demo →
platformCan I build against this from outside the UI? We already work in Claude and Cursor.
Over MCP, in both directions. Connect any MCP client over OAuth and it can search conversations, query pipeline through Opine SQL, and author agents, all scoped to what that person is allowed to see. There are no static tokens to manage. Opine is a client too, so the agents you build can call the third-party servers your org connects. One guardrail worth knowing up front: an agent authored over MCP saves disabled and can’t act until a human turns it on in the app.
evalsHow do I know an agent works before I put it in front of the org?
Run it against a dataset first. An evaluations workbench tests an agent against your data, so you find where its judgment goes wrong before your pipeline does. Every run keeps a record of its reasoning, its tool calls and its exact cost, and each agent journals its runs and distils them daily. Credits are prepaid with a per-run cost and an optional auto-refresh, so there’s no overage bill to discover later.
adoptionHonestly, our reps won’t maintain Salesforce. Does any of this depend on them changing?
No. That’s the design constraint. Fills draw from things that already happen: the calls, emails, Slack threads and meetings on the deal. Reps keep selling and the record catches up on a schedule. Every hygiene initiative that’s failed at your company failed because it needed somebody to type more. This one doesn’t.
governanceWe’re not comfortable with AI writing directly into our CRM. What are the controls?
Every capability is set on its own. Any write capability can pause for human sign-off before it acts, and field-level auto-apply is opt in, field by field. What we recommend and demo is read-only work automatic, writes gated, and gates relaxed as trust builds. CRM writes go through the acting user’s credentials, so your field-level security holds.
costWon’t running this across every deal get expensive?
Runs skip deals with nothing new to say. A run passes over deals with no material change since the last one, and unchanged values never trigger a CRM push. The scheduler does the cost control instead of you, and the bill tracks what actually happened in the pipeline rather than how many records are in it.
accuracyHow accurate is it? What happens when it reads a field wrong?
Every value carries its source, and there’s a known gap. Each suggested value cites the interaction it came from, so a review is a check against evidence rather than a judgment call. To be straight about the limit: we have an open report of agents misreading multi-select and checkbox fields as empty, and it’s being worked. Until it closes, leave those field types in review rather than on auto-apply.
setupShould we use scheduled field fill or an agent?
Same engine, different reach. Scheduled field fill is the workhorse for keeping configured fields current, each on its own cadence with its own review setting. Build an agent when the update is part of a bigger job, like sweeping active deals daily and updating what moved, or writing a full post-sales handoff into the CRM on Closed Won.
platformWhat is an AI context layer for RevOps?
The layer under your agents and your reporting. Opine is the AI-ready context layer for revenue operations. It reads the calls, email threads and documents your org already produces, normalizes them into one evidence-backed record per deal, and keeps that record current on a named cadence. Your dashboards, agents and warehouse all read the same record, which is what makes their answers accurate.