Alta

Breadth across several classes, with named systems in each. CRM is native and bidirectional across Salesforce, HubSpot and Pipedrive, with a Salesforce AppExchange listing packaging the agents for Agentforce. Data is drawn from more than 50 external sources including intent providers, job postings, news, product usage and engagement history.

Advertising is a write destination, with audience data pushed to Google Ads and Meta so paid targeting follows the accounts the agents are working. Luna posts insights into Slack, email or Teams and integrates with calendar tools, and messaging spans email, LinkedIn, SMS, WhatsApp and telephony. The vendor positions the platform to run on top of the existing stack rather than replace it, and the pricing page lists CRM and 50-plus tool integrations as included.

Sourcealtahq.com, altahq.com/ai and altahq.com/plansread 2026-09-02

Three specialized agents run one funnel rather than one agent running a step. Katie owns outbound: sourcing accounts from CRM and external signals, researching contacts, building lists and running multichannel sequences with condition-based branching that adapts in real time.

Alex owns inbound and calling: qualifying leads on arrival, holding an AI voice conversation, scoring intent, routing to the right representative and booking the meeting, and working existing accounts for cross-sell. Luna owns the intelligence layer: reading pipeline, surfacing recommendations and feeding targeting, timing and messaging back to the other two.

They share one brain, so signal from each channel changes what the others do, and the platform is marketed across six end-to-end motions from outbound pipeline through no-show prevention to closed-lost revival.

Sourcealtahq.com and altahq.com/airead 2026-09-02

Grounding on a maintained structure that combines the customer's own records with external signal.

The shared intelligence layer connects the CRM to more than 50 data sources (CRM activity, intent data, job postings, news, product usage and engagement patterns) and holds them as the standing basis every agent reasons from, rather than assembling context per request.

The vendor says every message draws from deal history, case studies, competitor intelligence and company context, so the customer's own material shapes what is said rather than a template being filled. The retrieval architecture itself is not documented.

Sourcealtahq.com/ai and altahq.comread 2026-09-02

Escalation is designed in, but nothing gates what goes out. Alex qualifies leads, scores intent and routes only ready buyers to the right representative, so the handoff to a person is a built mechanism rather than a policy, and the vendor says its agents are best used to augment rather than replace people, who keep discovery, complex conversations and closing. Granular access controls and hierarchical permissions limit who can operate the agents.

But no pre-send approval step, review queue, confidence threshold or hold is documented before an email, a LinkedIn message or an AI voice call reaches a prospect. That gap carries extra weight because Alex places autonomous outbound calls, a channel with its own consent rules. The trust page's line about your policies and your rules is posture rather than mechanism.

Sourcealtahq.com/faq, altahq.com/ai-inbound-agent and altahq.com/trustread 2026-09-02

Both certifications and access controls are documented. Alta is SOC 2 Type 2 and ISO/IEC 27001 certified, with documentation published through a trust portal at trust.altahq.com, plus annual penetration tests and bi-weekly vulnerability scans run by third-party security firms. Controls, all named on the trust page: SAML, granular access controls, hierarchical permissions, an audit log, IP restrictions and a single-tenant data warehouse.

Infrastructure runs on AWS managed services isolated in a private VPC with encryption at rest and in transit by default. Earlier checks also found a published subprocessor list with 30 days' notice on changes, data subject export or erasure within 7 days, and a self-serve DPA in the admin portal; these were not re-confirmed on the latest read.

Sourcealtahq.com/trust and altahq.com/faqread 2026-09-02

Two records, neither of them a trace of the agent's reasoning. Account activity: an audit log is named among the controls on the trust page, alongside granular access controls and hierarchical permissions, so who did what inside the platform is recorded. Business state: Luna monitors pipeline in real time, surfaces recommendations, detects patterns and alerts the revenue team to material changes in Slack, email or Teams, and consolidates reporting across CRM, ad platforms and data sources.

What is observable is the pipeline and the account, not the agent: no decision trace, reasoning log or per-run record of why Katie chose a channel, why Alex scored a lead as qualified, or what changed between two campaign attempts is documented. The homepage animates such a surface, showing a channel switch and its measured effect, but that is marketing rather than documentation.

Sourcealtahq.com/trust and altahq.com/airead 2026-09-02

No memory construct is documented. Luna feeding Katie and Alex better targeting, timing and messaging as it learns from every interaction, and call outcomes feeding back into targeting, is an optimization loop, covered under testing. The agents otherwise draw on the shared intelligence layer over the CRM and more than 50 data sources, and a CRM record is not memory. No per-account or per-conversation memory, retention period, inspection or deletion path is described on the agent pages, the FAQ or the trust page.

Sourcealtahq.com/ai and altahq.comread 2026-10-01

A real tenancy control, but no choice of location. The trust page lists a single-tenant data warehouse among the controls a customer gets, so the customer's data can sit in a dedicated store rather than a shared pool.

What is missing is customer control over where the software runs, through region selection, a VPC, on-premises or the customer's own tenant: the single tenancy is Alta-operated, the private VPC is Alta's own architecture rather than a customer deployment, and no selectable region or EU-hosted option is offered.

The customer FAQ has stated that data is stored in US data centers, while the published subprocessor list named AWS Frankfurt alongside Oregon, an inconsistency that matters to EU customers. Either way, no region is customer-selectable.

Sourcealtahq.com/trustread 2026-09-02

A named set of prebuilt agents the buyer selects among. The home page lists Katie, AI Outbound Agent ("Build pipeline, automatically"), Alex, AI Inbound Agent ("Qualify every lead, instantly") and Luna, AI Growth Agent ("Orchestrate your full GTM"), and Katie and Alex each have their own product page. Katie and Alex are whole products doing their own jobs, outbound pipeline and inbound qualification, and either stands without the other; Luna is the intelligence layer above them. The quote form asks which motions the buyer wants to run, so the agents are chosen rather than internal roles.

Sourcealtahq.com, altahq.com/ai, altahq.com/ai-outbound and altahq.com/plansread 2026-10-01

Both intake routes are covered. Outbound starts on signals: buying signals, engagement patterns and trigger events drawn from more than 50 data sources, including intent data, job postings, news and product usage, determine when outreach begins, and the vendor's example of a champion changing employer shows the granularity claimed.

Inbound is real-time: Alex answers a new lead the moment it lands, with a sub-30-second response time published against a stated industry average of 42 hours, qualifying by AI voice call or chat and routing to a rep's calendar. Channels span email, LinkedIn, SMS, WhatsApp, telephony and chat, sequenced with condition-based branching that adapts in real time, plus ad audiences as a write destination.

Sourcealtahq.com and altahq.com/ai-inbound-agentread 2026-09-02

No model is named and the customer has no choice. Nothing across the agent pages, the trust center, the FAQ or the pricing page offers a model selector, per-agent model choice, routing policy, bring-your-own-model or bring-your-own-key, and no provider is identified anywhere.

The trust page's line about full control over AI models and data sharing (your policies, your rules) sits under a heading about Alta's own in-house security governance and reads as Alta controlling its models rather than the customer selecting one.

The vendor's description of a dedicated team overseeing security and model development points the same way, toward proprietary or internally managed models.

Sourcealtahq.com/trust and altahq.com/airead 2026-09-02

A programmatic way in exists, but there is no public build surface. The site's navigation and footer carry three agent pages, integrations, trust and pricing, four solutions pages, about, careers, contact, press, comparisons, and resources covering blog, glossary, FAQ, ebooks, a help center at support.altahq.com, a partner program and a learning academy.

There is no developer portal, no API or docs subdomain, no published API reference, no SDK and no MCP server, so extension runs through prebuilt connectors rather than an open build surface. Alta's security documentation names API and OAuth ingress alongside CSV and web forms, a documented programmatic route in, though that page was last read on 30 July. A Salesforce AppExchange listing packages Alta for Agentforce, which is Alta used inside another platform.

Sourcealtahq.com and altahq.com/integrationsread 2026-09-02

An optimization loop after deployment, aimed at the outreach rather than at the agent. The Luna page says "Luna runs A/B tests, spots patterns, and self-optimizes", improving messaging, timing and targeting with every interaction, and that it detects what is converting across thousands of interactions so the agents act on what works.

Nothing documents how variants are chosen, what a test reports, or where a revenue leader reviews results. Nothing lets a customer test an agent before it reaches a prospect either: no sandbox, simulation, dry run or scored evaluation. What gets tested is the customer's messaging.

Sourcealtahq.com/airead 2026-10-01

Every documented channel runs through a programmatic route. The calling agent places telephony calls, LinkedIn, email, SMS and WhatsApp operate through integrations, and the CRM connections are native bidirectional syncs, so nothing describes an agent driving an interface the vendor does not control. No browser control, headless session, virtual desktop or screen operation appears on any page.

Sourcealtahq.com and altahq.com/integrationsread 2026-09-02