Ask your business systems anything. No data lake, no pipelines.
SourceLace connects your AI assistant to your CRM, ERP and other business systems, live and with each person's own permissions. Data is read where it lives instead of copied into a lake, and every action is audited.
Built on the open MCP standard, for Claude, ChatGPT, Cursor, VS Code or your own agent. Or use the SourceLace app, with charts and Word and PowerPoint reports.
Your answers live in five systems. Today there are three bad ways to reach them.
Move it all to a data lake
Months of pipelines and cost, and the data is stale and exposed the day it lands.
Give the AI an admin login
Everyone sees everything, and nobody can tell who did what.
Build one-off integrations
Each one breaks with the next schema change or new custom field.
One URL in your AI app. Each person signs in as themselves.
Connect once
Add SourceLace to your AI app with one address. Each person signs in to Salesforce with their own account, in their own browser. Per-person sign-in for SAP is coming soon.
Ask in plain language
The assistant finds the right objects, including your custom ones, queries them live, and joins Salesforce and SAP results in memory.
Results don't linger
Query results are held in memory for up to 30 minutes, never written to a database, and then they're gone. No lake, no sync jobs to babysit. Chats you save in the SourceLace app, with the rows they show, are encrypted and deleted on your schedule (30 days by default).
The source system still decides. SourceLace never widens access.
- Each person's own permissions. Salesforce profiles and sharing rules apply exactly as they do in Salesforce. SAP authorizations will apply the same way once per-person SAP sign-in ships.
- Only what you keep is stored. Query results are held in memory for up to 30 minutes and never written to a database. What you keep in the SourceLace app is stored, encrypted with your organization's key: saved chats (which can include rows you were shown) and the files made in them are deleted after your retention period (30 days by default), and skills stay until you delete them.
- Writes you approve. Off by default and enabled per object by an admin. Every change is previewed field by field and applied once, only after the person confirms.
- Audit trail. Who asked what of which system, and what changed. Metadata only, never the data. Admins search and filter it in the app.
- Encrypted credentials. Sign-in tokens are encrypted with a key per customer. For most systems people sign in on the system's own page, so SourceLace never sees their passwords. Databases have no such page, so each person types their own database login into SourceLace, which keeps it encrypted and never logs or shows it.
- Bring any AI. Use any MCP-capable app; SourceLace then sends nothing to an AI model itself. The SourceLace app's assistant uses Claude, or your organization's own AI key.
- Cloud or self-hosted. Run it in our cloud today. Running it in yours, from the same codebase, is coming soon.
| Status | New | Closed |
| Reason | (empty) | Duplicate |
From answers to action: agents that do the routine work, within your rules.
Agents run on a schedule or when something happens in a system, using the same governed access as a person: their owner's permissions, every step audited.
Before the business review
Two days before each QBR on the account owner's calendar, build the deck from CRM, ERP and support data and leave it for review.
When a severity 1 case lands
Draft the escalation email to the account team with the account's open orders, renewal date and recent cases, ready to send.
When health drops
Recalculate the account's health score, propose the update, and open a follow-up task for the customer success manager.
- Drafts run on their own. Emails and decks are prepared for a person to review and send.
- Changes wait for approval. Record updates are previewed and applied only after a person approves, unless an admin allows that one action to run by itself.
- Never more access than their owner. Each agent is limited to the sources, objects and fields an admin allows, with run limits and a stop switch.
Where we're going: agents for every team that runs on business systems, then voice, so people can ask and act by talking, and later customer-facing service built on the same governed foundation.
Standard and custom objects, from day one.
Questions that used to take a week of exports.
- "Give me a 360 of Corvanta Corp before Thursday's QBR: open deals, cases, contacts and risks."Salesforce · one record with its related lists
- "Which renewals this quarter have overdue SAP invoices?"Salesforce + SAP · joined in memory
- "Close the six duplicate cases on Corvanta Corp."previewed, then confirmed by you
Run a 30-day pilot on your own org.
We connect your Salesforce org, with your permissions and your AI app, so your team can try real questions on real data. Tell us which other systems you use; we're adding them in that order.