Every call is an MCP tool call
Outbound campaigns, AI voice agents, and voicemail triage are all MCP tools — on Twilio, SignalWire, Plivo, Vonage, or your own SIP trunk.
See voiceDecisions, not dashboards.One system, not a stack.
Most tools give you context derived from abstraction. Pact gives you abstraction derived from context: it reads Pact reads the actual records, calls, and history first, then tells you what they mean, with the metrics cited underneath. Ask for it in a sentence — out loud, if you like.
Try it now
Your turn — ask it anything.
This is the live agent, answering from a demo workspace. No signup, no download.
Pact: Hey — I'm Pact. Ask me about your pipeline. Try: which deals are slipping this week?
Call +1 (424) 722-8627
This browser can't do speech recognition, so the phone line is the demo. Same agent, same data.
Context from abstraction
A dashboard is an abstraction of what somebody retyped.
Most tools give you context derived from abstraction: they start at a dashboard and let you drill toward the facts. But the facts were never there. The people closest to the customer spend the day re-entering what already happened, so every number built on those fields is a negotiation about whose export is right, while the thing that actually mattered stays in the call, the thread, and the hallway.
Call · Tuesday
“Security review has to clear before we can sign anything.”
Email thread
“Looping in our new CFO — she owns the budget now.”
Slack, internal
“They went quiet after the pricing call. Not sure we're still in it.”
none of it reached the record
- Stage
- Proposalunchanged 41 days
- Next step
- —
- Notes
- lm vm
- Last updated
- 41 days ago
Abstraction from context
Abstraction derived from context. The record comes from the conversation.
Pact does the opposite. It reads the actual records, calls, and history first, then works out what they mean. Speech is the fastest interface a rep has, and a language model reads a transcript better than a form ever could, so people talk and the system does the structuring: what moved, what stalled, what is slipping. The words it came from stay underneath the result, so you can check the work.
DMLive call · 14:02
“Just wrapped the Pinecrest call — they’re moving to proposal, targeting Oct 21 close, eight forty.”
Pact captures
Pinecrest Retail Group — Q4
Deal · updated from the call
- Stage
- Proposal
- from “moving to proposal”
- Value
- $840,000
- from “eight forty”
- Close date
- Oct 21
- from “targeting Oct 21 close”
Consent-checkedAudit-loggedNo form opened
What Pact does
Context from abstraction produces questions. Abstraction from context produces decisions.
Open Pact and the day is already ordered: what is stuck, what is waiting on you, who needs an answer, each conclusion with the metrics cited underneath it. Ask what is slipping and it answers from your own records, out loud or in text. Dictate an update and it lands on the right record, consent-checked on the way out. There is no dashboard to interpret and no form to fill in.
Pact is pre-launch. Both screenshots above are the product as it runs today, captured from a live workspace — not a mockup and not a roadmap. You can talk to the same agenttalk to the same agent at the top of this page without signing up.
Time to operation
A finding never ships without a fix attached.
Most platforms hand you a configured empty box and a documentation site. In Pact the thing that needs doing shows up as a button, with the reasoning and the proposed change already attached: re-date the deal, re-confirm the consent, re-engage the champion. Tap it and you land on the record, not on a to-do. An alert with no action is just another task, and you have enough of those.
Deal · close date lapsed
Close date passed 23d ago — re-date or advance
Stage Proposal, unchanged 41 days
Last touch was the pricing call
Move close to next quarter, or advance to Negotiation
Re-date or advanceConsent · expiring
Re-confirm consent — expires in 12d
Email opt-in from the June webinar ages out
Two live sequences depend on it
Re-confirmation request, consent-checked before it sends
Re-confirmChampion · silent
Re-engage Dana Whitfield — champion silent 19d
No reply since she asked for the security packet
Deal is the largest in the quarter
Open the thread with the packet attached
Re-engageWhere the answer came from
The number we could not read is not a zero.
Every part of an answer is shown at the level of evidence behind it: a row from your workspace reads as a row, a computed number arrives with its arithmetic, and the model’s reading is visibly a reading. When something could not be read at all it appears as a stated gap naming what failed — not a zero, not a blank, and not a confident sentence covering the hole.
Asked: how did pipeline move this quarter, and which reps drove it?
$2.41M created this quarter
38 deals in your workspace
Read from your workspace.Up 18% on last quarter
2.41M ÷ 2.04M − 1
Computed, shown with its arithmetic.“We want the security review finished before we talk about renewal.”
Call transcript · 4 September
Quoted from a source, named.Most of the quarter’s lift is mid-market: the three largest new deals all came from the segment you started sequencing in July.
The model's reading, not a record.Win rate by rep — not established
The per-rep rollup did not return. Reported as a gap, not as 0%.
Not established. The gap is the answer.One system
The stack a business assembles in its first year. In one place.
Six to ten vendors means six to ten contracts, security postures, and compliance boundaries composed into something that reads as coherent and does not operate as one. When it breaks across a seam, nobody owns it. Pact is one vendor, one boundary, one system that can see the whole picture.
Instead of
A CRM
Accounts, contacts, deals, and pipeline
Instead of
Marketing automation
Sequences and journeys, consent-checked per send
Instead of
A business phone system
Inbound and outbound calling, consent captured on the call
Instead of
Call transcription
Transcripts, summaries, and next steps from every call
Instead of
Consent management
A per-channel, jurisdiction-aware consent ledger
Instead of
Quoting software
Quotes and orders on the same deal record
Instead of
Data pipelines
Imports and ETL into one tenant-isolated store
Instead of
An AI assistant
Hey Pact — ask over live records, by voice or text
Instead of
Agent tooling
An MCP server with 180 first-party tools
Instead of
Task and calendar tools
Tasks and calendar on the same records
Every tile is a module running in the live workspace the demo on this page answers from — not a roadmap item. What the product cannot do yet is left off rather than dressed up.
How Pact works
One system. Three moves.
Not ten features to learn. One place to centralize, act, and see.
Import from a CSV or a live source, and every duplicate collapses onto one record you can export in full at any time.
The consent check runs at send time, per channel. A withheld contact is skipped and the skip is written to the audit trail.
Because reporting reads the same rows the app writes, a number on a dashboard and a number in a record cannot disagree.
Don't take our word for it — touch the real product below.
How Pact works
One platform, four pillars.
The substrate every revenue team already pays four vendors to imitate.
One source of truth
Accounts, contacts, deals, signals, and tracking events live in a single tenant-isolated store.
Sequences + playbooks
Multi-channel cadences and event-driven playbooks, shipped behind feature flags.
Consent-first by default
GDPR / CCPA / CASL state per-channel per-contact with an immutable audit log.
Real-time analytics
Pipeline velocity, ICP distribution, open rate by provider, AI cost ledger — queryable in-product.
Model Context Protocol
180 tools your AI can already use.
Pact ships a native MCP server. Point Claude, Cursor, or any MCP client at your workspace and it can read your pipeline, run an agent, place a call, or write back a note — every call consent-filtered, audit-trailed, and cost-attributed per tenant.
CRM
Read your pipeline
Query accounts by any field — owner, stage, ICP tier, open value.
query_accounts(
stage="Negotiation",
min_open_value=25000,
limit=3
)[
{ "name": "Acme Corp", "stage": "Negotiation", "open": 50000 },
{ "name": "Northwind Ltd", "stage": "Negotiation", "open": 38000 },
{ "name": "Globex Inc", "stage": "Negotiation", "open": 27500 }
]Illustrative preview with synthetic data — not a live call. Every real invocation is consent-filtered, audit-trailed, and cost-attributed per tenant.
Bring your own keys
Connect your own provider keys in-product and pay them directly. Nothing is hardcoded; nothing routes through Pact's account.
OpenAIAnthropicTwilioSignalWirePlivoVonageDeepgramAssemblyAIPostmark+ your SIP / LLM endpoint
How BYOK works Consent-native, not bolted on
Retrieval is consent-filtered before a record enters context, and your data exports in full on any plan. SOC 2 Type II audit is underway (2026-Q4).
Trust & securityMCP depth
Every CRM has an MCP server now. Most of them can only read.
An agent that can only read is a search box with extra steps. The column that matters is the third one — whether the agent is allowed to change anything.
| Vendor | MCP tools | Agent can act? | Surface |
|---|---|---|---|
| PactSource | 180 | Yes — read and writeAn agent can create, update, and act on CRM records through the same tools it reads them with. | CRM records |
| Gong1Source (opens in a new tab) | 3 | No — read-only“The MCP server is read-only and does not create, update, or delete data in Gong or the CRM.” | CRM records |
| Salesforce2Source (opens in a new tab) | Not public | Yes — read and write“Read, create, update, and delete Salesforce records with tools that respect field-level security and sharing rules.” | CRM records |
| HubSpot3Source (opens in a new tab) | Not public | Yes — read and write“enables you to give any MCP-compatible AI tool or agent secure read and write access to your HubSpot CRM data” | CRM records |
| Attio4Source (opens in a new tab) | Not public | Yes — read and write“Read operations: Auto-approved… Write operations: Request user confirmation” | CRM records |
- Footnote 1. Gong: Three tools: ask_account, ask_deal, generate_brief.
- Footnote 2. Salesforce: Salesforce ships several MCP servers. The separate DX server advertises 60+ tools, but those are developer tooling (build, test, deploy) — not CRM records. Hosted-server tool count is not published.
- Footnote 3. HubSpot: HubSpot documents a tool list but states no total; we do not estimate one.
- Footnote 4. Attio: No published total; two reads of the docs page yielded different counts.
Vendor-published facts, captured 2026-07-31. Tool count alone is not a quality measure — what matters is whether an agent can act, not just read.
List price per seat
- SalesforceSales Cloud Enterprise
- $175/ seat / monthPricing page (opens in a new tab)
- HubSpotSales Hub Professional
- $90/ seat / monthPlus a one-time $1,500 onboarding fee — mandatory, and not included in the per-seat price above.Pricing page (opens in a new tab)
- AttioPro
- $79/ seat / monthPricing page (opens in a new tab)
Annual-commit list price from each vendor's own pricing page on 2026-07-31. HubSpot additionally requires a mandatory one-time onboarding fee of $1,500 on Sales Hub Professional — the line most comparison tables leave out, and the one that moves a first-year number most.
Don’t take our word for it. Ask it something.
A real Pact agent on the same 180 MCP tools the product runs on. No signup, no card — it answers out loud.
Live agent · demo data
Hey — I'm Pact. Ask me about your pipeline. Try: which deals are slipping this week?
Or tap a question
Isolation
One workspace cannot read another’s rows.
The claim every buyer tests first. Here is exactly where it stands, including the part that is not finished.
Five cross-workspace reads found and closed
An accounts list and its count, a stage-history lookup, an automatic merge, four customer-health reads, and pipeline after a data migration.
One unauthenticated endpoint deleted, not patched
A live ticker could return data across workspaces. It had no owner and nothing depended on it, so the safest version was the one that does not exist.
Isolation policies on every workspace-scoped table
783 of 783, with the workspace identity bound per transaction rather than on a connection a pooler might hand to someone else.
Staged, not live — Making the database itself refuse a query that forgets is built and staged, and is not switched on today. We will say so here on the day it is.
The arithmetic
What after-call admin costs your team, in hours.
Pact's voice agent writes the call back to the CRM itself. Set the four numbers below to your team and the model shows the time that returns to selling.
Your team
8 AEs
1200
12 calls
160
$15,000
1,000250,000
2%
Per CALL, not per qualified deal — outbound teams are typically low single digits. If you track win rate on deals, that number is much higher and does not belong here.
125
Returned to selling
32hours / week, across the team
8 AEs × 12 calls × 4 minutes of admin × 5 days. That is the whole calculation — no assumption about your team beyond the four numbers you set.
Modeled revenue effect
$757,029/ year
If that recovered time goes back into calls at your existing rate, it is 2,523 more calls a year, each valued at the 2% per-call win rate you entered — unchanged — times your average deal size. It is a model of what the returned time is worth, not a forecast of what you will book.
Check the model against yourself. The same arithmetic says your team already books $6,624,000 a year (22,080 calls × 2% × $15,000). If that is far off, the win rate you entered is an opportunity rate rather than a per-call rate, and the figure above is overstated by the same factor.
How this is calculated
- Admin per AE per day = 12 calls × 4 minutes = 48 minutes.
- Hours returned per week = 8 AEs × 48 minutes × 5 days ÷ 60 = 32 hours.
- Share of the selling day = 48 minutes ÷ 420 minutes = 11.4%.
- Additional calls per year = 8 AEs × 12 calls × 5 days × 46 weeks × the share above = 2,523 calls.
- Modeled revenue = 2,523 calls × 2% win rate × $15,000 = $757,029.
Assumptions
- Assumes 4 minutes of after-call admin per call — call logging, field updates, and drafting the follow-up — which Pact automates.
- Assumes a 7-hour selling day and a 5-day selling week.
- Assumes 46 selling weeks a year, leaving room for vacation, holidays, and training.
- Assumes recovered time is spent making calls at the same rate the team already makes them — no productivity bonus is applied to the extra time.
- Assumes no change to win rate. The revenue line applies the win rate you entered, unchanged, to the additional calls — we model no uplift, because we have no basis for one.
- Values each additional call at your win rate × your average deal size. That means the win rate here is per call, not per deal: if you track win rate on qualified deals, enter the lower per-call figure or this line will overstate the result.
- This is a model, not a promise or a forecast. It contains no customer data, because Pact has none to draw on yet.
The difference
Principles, not fine print.
Three commitments that hold on every plan — no asterisk, no upsell.
Zero lock-in
Export your entire workspace on any plan, including Free — records, activity, and consent state, in one archive with a manifest and your own audit log inside. Your data is never held back.
AI you control
Bring your own LLM provider, encryption keys, and email sender. The AI runs on infrastructure you own, and every call is consent-filtered and cost-attributed.
Everything is logged
Every action writes to a real-time event stream and an append-only audit trail — a complete, tamper-evident record of who did what, and when.
Trust
Built for the procurement question, not just the demo.
Honest about what's certified today and what's in flight. Pact's own SOC 2 Type II audit is underway — the substrate it runs on carries Type II already.
SOC 2 Type II
In progressPact's own audit — substrate provider carries Type II today
Target 2026-Q4
GDPR
CompliantPer-channel consent state + immutable audit log
ISO 27001
In progress93 Annex-A:2022 controls + Statement of Applicability, tracked in-product
HIPAA BAA
CompliantAvailable on Team and Enterprise plans
CCPA
CompliantConsumer-rights workflows + DSAR export
BYOK encryption
CompliantPer-tenant data keys wrapped by your KMS
Uptime
0.0%
Platform availability target. Live per-component 90-day numbers are on the public status page.
See live status →Security pack
Architecture, controls matrix, sub-processors.
The same answers we hand to enterprise procurement teams — live on the Trust Center and the security page, no email gate.
No lock-in
Your data is yours. On every plan.
The things teams get burned on — export, keys, audit — are commitments here, not upsells.
Zero lock-in
Export your entire workspace on any plan — including Free. Records, activity, and consent state in one archive, with a manifest and your own audit log inside. Your data is never held back.
AI you control
Bring your own LLM provider, encryption keys, and email sender on the Team plan and up. The AI runs on infrastructure you own, and every call is consent-filtered.
Everything is logged
Every action writes to a real-time event stream and an append-only audit trail — a tamper-evident record of who did what, and when.
Four contracts, or one.
Swipe through the four questions a patchwork makes hard to answer.
One customer
Four separate tools: A different copy of them in each tool, reconciled by hand.
With Pact: One tenant-isolated record that CRM, marketing, and success all write to.
Can we email them?
Four separate tools: Ask the consent vendor, then hope the sending tool agrees.
With Pact: Consent state per channel, checked before the send leaves.
Where do numbers come from?
Four separate tools: Three exports and a warehouse job to make them agree.
With Pact: Pipeline, marketing, and success reporting off the same live data.
Getting your data out
Four separate tools: An upgrade, a support ticket, or a migration project.
With Pact: A full export, manifest included, on every plan — including Free.
GDPRCCPABYOK encryptionSOC 2 (in progress)
Built for the whole revenue org
20 roles, one system.
Each one has a page showing exactly what it gets. Same record underneath.
Sales · Marketing · Customer Success · Operations · Trust & Security
See what each role getsFor your role
Built for every role in the revenue org.
One platform, twenty doors. Find the workflow that talks like you do — every page links to the shipped surfaces behind the claims.
Sales
- Pact for SalesCRM, sequences, dialer, and an AI inbox in one keyboard-driven app.See the page
- Pact for SDRs & BDRsOutbound cadences, prospecting signals, and booked meetings without the tab sprawl.See the page
- Pact for Solutions engineersRFP autofill, evidence vault, architecture proposals, and demo sandboxes on demand.See the page
- Pact for Sales OpsDeal desk, CPQ, comp, and forecast hygiene on one event-sourced ledger.See the page
- Pact for VP Sales & CROsPipeline health, a forecast that doesn't drift, and team performance in one view.See the page
Marketing
- Pact for MarketingCampaigns, segments, and consent on one substrate — no nightly sync.See the page
- Pact for Marketing OpsLifecycle journeys, attribution, and deliverability that survive a reorg.See the page
- Pact for Growth & LifecycleActivation, nurture, and retention loops that fire on real product events.See the page
- Pact for Customer MarketingExpansion campaigns, advocacy from your promoters, and stories that source themselves.See the page
Customer Success
- Pact for Customer successHealth scores you can debug and renewals you can defend.See the page
- Pact for Support teamsEvery channel in one queue, with MCP-native tickets your AI can resolve.See the page
- Pact for Voice-first teamsBring your own carrier; every call transcribes, coaches, and files itself.See the page
- Pact for CX leadersCSAT, NPS, SLA, and health across support and success in one scorecard.See the page
Operations
- Pact for RevOpsOne source of truth for pipeline, attribution, data, and consent.See the page
- Pact for Data teamsYour CRM as an event-sourced warehouse source, not a snowflake.See the page
- Pact for Finance leadersForecast the quarter from the same ledger the AE moves — AI spend capped.See the page
- Pact for IT & AdminSCIM, RBAC, SSO, BYOK, and an immutable audit log without a ticket queue.See the page
- Pact for FoundersFrom first design partner to first hire on one stack — no re-platform.See the page
Architecture
What it's actually built on.
No architecture slide with a box for something we haven't written. Every layer below is running in production right now, and the provider tier runs on your keys, not ours.
The deploy pipeline, DNS cutover, and per-component status are public. See live status · Security overview
What's new
Shipped recently.
Every card links to the change itself, not to a description of it — and nothing is listed here that we could not check against production first. Most recent: September 15, 2026.
- FixedIntegrations & API
Six invented numbers removed from Pact Data, and the identity graph learns to report itself
Pact Data's profile and vault pages now show your workspace's own numbers — and say plainly when a figure could not be read, instead of showing a zero.
Read more - Fixed
The health refresher stops writing snapshots for companies no tenant owns
Fixed a defect where customer-health scores could be computed and stored for accounts that belong to no workspace, under every workspace's identity.
Read more - New
Health and churn carry their age, so a stale score cannot read as current
Health scores and churn predictions now show how old they are. A score nobody has recomputed in months no longer looks like this morning's.
Read more - NewData & analyticsIntegrations & API
The join everyone documented as impossible is one hash away
The join everyone documented as impossible is one hash away
Read more - Fixed
The bounded wait for a persisted relay turn could not wait where it was called
The bounded wait for a persisted relay turn could not wait where it was called
Read more - NewIntegrations & API
A notification names who, what and where — and the link it already carried gets used
A notification names who, what and where — and the link it already carried gets used
Read more
Frequently asked
The questions every team asks first.
Pricing, security, migration, and the “is this for us?” question. Pricing mechanics live in detail on /pricing.
How does pricing work, and is there a trial?
Sign up and the first 14 days run on a full Pro trial — no credit card. After that it's per-seat monthly on Pro and Team, usage-aware so contact and event caps grow with the plan. If you don't upgrade by day 14 the account drops into the Free plan (1 user, 500 contacts, 1k events / month); nothing breaks and no data is re-migrated. Full breakdown plus an interactive calculator is on the /pricing page.
Can I import from another CRM?
Yes. The live Salesforce, HubSpot, Pipedrive, and Zoho migrators move accounts, contacts, deals, owner mapping, and consent state in one guided in-product run, and the CSV importer covers most other sources. Close and Apollo importers are next. White-glove migration help is included on Team and Enterprise.
Is my data secure, and where does it live?
Every workspace is tenant-isolated — your data shares no schema, no row, and no encryption key with any other tenant, and row-level isolation policies now cover all 783 workspace-scoped tables. Five specific reads that could have crossed a workspace boundary were found and closed this month, and an unauthenticated endpoint that could return data across workspaces was deleted rather than patched. The switch that makes the database itself refuse a query that forgets is built and staged, and is not switched on today; we print that on the homepage rather than let a policy count read as a guarantee. The data layer carries SOC 2 Type II and HIPAA-ready attestations today; Pact's own SOC 2 Type II audit is in progress with a 2026-Q4 target. Every tenant is served from US-East today. Per-tenant region pinning (US-West, EU-West, Asia-Pacific) is built and recorded in the platform, but the regional deployments are not live yet and enforcement is not armed — so we do not offer it as a commitment; ask us if you need a region and we will tell you where it actually stands. Self-hosting in your own VPC is an Enterprise conversation — same container images, scoped during onboarding.
Do you integrate with the tools we already run?
73 connectors in the marketplace — Slack, MS Teams, Stripe, Resend, Postmark, SendGrid, Amazon SES, OpenAI, Anthropic, AWS Bedrock, Snowflake CDC, BigQuery CDC, and more — each labeled with its current sync depth. Anything not on that list is covered by a full REST API with a typed OpenAPI schema, webhooks for every domain event, and a Postgres-backed event stream you can subscribe to with standard CDC tools. Public API docs live at /docs/api; webhooks are configured per tenant in the admin UI.
How long does it take to get value?
You can import your first contact list and ship your first sequence within an hour of signing up — the Salesforce migrator runs a full guided import in a morning. Replacing an entire stack end-to-end is a 6–12 week project, and white-glove help is included on Team and Enterprise.
What happens to my data if I cancel?
Full export of every contact, deal, sequence, consent record, and audit event in standard formats (CSV + JSONL), available on every plan including after cancellation. Read access stays on during the 30-day grace window. After grace, the data is permanently deleted.
More questions — mobile, plan limits, white-labeling — see the full FAQ.
Full access, 14 daysNo card required.
Start free