Lightfield

Lightfield is an AI-native CRM designed to build and maintain its customer model from the work a team is already doing rather than relying on manual record hygiene. It connects calls, email, meetings and other customer signals, preserves a running memory of each relationship, and uses that context to answer questions, update records and perform sales work.¹²

The architecture is more consequential than the usual “AI assistant inside a CRM” pattern. Current product material describes customer memory as a versioned source of context that agents can reason over, while VentureBeat independently describes a schema-flexible system that stores rich interaction history and extracts structure from it as needed.¹³⁴ That allows fields, summaries and classifications to evolve without requiring the company to decide every part of its CRM schema before the system has learned how the business actually sells.

Lightfield’s chat layer can query and change CRM data, prepare meetings, summarize conversations and surface customer patterns with citations. The product has also moved into more active workflow execution: automations can run agent-driven tasks, sequences can draft or send outbound communications subject to configured review, dashboards expose operating metrics, and recent releases added document, spreadsheet and presentation creation inside chat.¹ The product changelog shows unusually fast expansion through mid-2026, including dashboards, Slack and Microsoft Teams work, LinkedIn messaging, sequence sentiment, field-history tooling, model selection for automations and broader direct table editing.¹

This breadth is paired with explicit human-review concepts. VentureBeat’s launch coverage reports that Lightfield acknowledged hallucination risk and described human approval for consequential customer communications and critical updates.³ Current product history likewise includes “For review” workflows for agent-drafted work. That is the right design direction for a CRM whose mistakes can affect customers, pipeline and revenue.

The integration surface is becoming substantial. Lightfield has native synchronization for mail, calendar, Slack and LinkedIn on current plans, a REST API with programmatic access to core CRM objects and conversation data, HTTP/webhook support from workflows, and an MCP endpoint using OAuth 2.1 for agent clients such as Claude Code, Cursor, ChatGPT and Codex.⁵⁶⁷ Slack access follows the user’s existing Lightfield permissions rather than creating a separate unrestricted data surface.⁸

There are still meaningful integration gaps. Lightfield’s own API documentation says direct CRM-to-CRM integrations are not currently the primary path; teams can run systems in parallel or build synchronization themselves.⁶ For organizations with a deeply customized Salesforce or HubSpot estate, the migration and coexistence burden therefore deserves explicit testing rather than assuming that an AI-native data model eliminates integration work.

The live pricing page currently uses a credit-based model with unlimited seats and all core features on every standard plan.⁵ Starter is pay-as-you-go with 5,000 free credits and a $0.040 per-credit rate. Pro is $1,000 per month with 400,000 credits per year at a $0.030 reference rate and adds an 8-hour support SLA, dedicated customer-success support, migration help and assistance configuring the data model and automations. Growth is $2,000 per month with 850,000 annual credits at $0.028 per credit and adds forward-deployed GTM, outbound campaign setup, ICP/target-account configuration, campaign design, inbox warming and performance check-ins.⁵

The usage economics are important because the headline subscription price is not a complete cost measure. Lightfield recommends budgeting roughly 37,000 credits per month for each active seller whose inbox, calendar and calls run through the platform. The company translates that to about $1,450 per seller per month on pay-as-you-go or roughly $1,100 on an annual commitment.⁵ Overage does not stop usage; extra credits are billed at the plan rate. Unused annual credits do not roll over at year end.⁵

That can make Lightfield attractive as a consolidation play if it genuinely replaces a CRM, meeting recorder, conversation-intelligence tool, outbound tooling and portions of RevOps work. It can also become expensive quickly when a team has several heavily active sellers. Buyers should model cost from representative inbox/call/automation volume rather than compare only the $1,000 Pro sticker price with a conventional CRM seat.

There is also a material documentation inconsistency that buyers should resolve before signing. The live pricing page says every plan includes unlimited seats and uses the credit structure above.⁵ Yet a Lightfield support article updated July 30, 2026 still describes Starter at $89 per user per month and Pro at $899 per workspace per month plus paid Full and Capture seats.⁹ The April 22 Terms likewise describe monthly or annual “Subscription Seat Fees” rather than the current unlimited-seat credit model.¹⁰ The live pricing page is the clearest current sales surface, but two other current first-party documents materially conflict with it. Contract and order-form economics therefore need to be checked directly rather than inferred from any single public page.

Independent exact-product evidence is still concentrated. VentureBeat has published two substantive pieces. Its November 2025 launch coverage describes the product’s automatic capture, unstructured relationship history, schema flexibility, natural-language querying and AI-generated sales work, while also discussing privacy and hallucination risks.³ A May 2026 CRM-category article again characterizes Lightfield around an agent-native architecture, automatically captured interactions and a queryable memory model rather than treating AI as a feature layered on a conventional record database.⁴

Those articles corroborate the architecture, but they do not amount to a controlled reliability or ROI study. Several adoption and outcome numbers in the launch article come from Lightfield or named customers rather than an independent dataset. They are useful examples, not a basis for assuming similar results across buyers.³

Product Hunt adds a small qualitative user signal. The current page shows roughly 2,100 followers, a #2 Product of the Day launch and one visible customer review praising automatic capture of prospect needs and the usefulness of the resulting customer context outside sales operations.¹¹ It does not expose a reliable aggregate Lightfield rating/sample large enough to use as a satisfaction input. Launch points and followers are adoption/attention signals, not quality scores.

Security documentation is comparatively strong for a young product. Lightfield states that it has SOC 2 Type II audits, supports HIPAA through a BAA, addresses GDPR through its DPA and SCCs, and encrypts data in transit and at rest.¹²⁵ Enterprise-only controls on the current pricing page include role-based access control, SSO and BAA availability.⁵ The pricing page currently labels ISO 27001 as “Soon,” so it should not be treated as a completed certification based on more general marketing language elsewhere.⁵

The privacy and legal footprint is substantial because the product works precisely by ingesting sensitive business context. The Privacy Policy says Magical Tome, Inc. dba Lightfield may process account information and personal data contained in user content, including files, documents, images, audio and video, as well as technical/usage data.¹³ The Terms state that customers retain ownership of uploaded files but grant Lightfield a license to use them to provide the service and for internal research and development such as improving the service; Lightfield may also derive deidentified or aggregated data.¹⁰

The Terms also contain two caveats that matter operationally. First, users are responsible for obtaining appropriate consent for recordings and transcriptions. Second, deletion may not always remove every copy of Customer Files immediately from all service systems.¹⁰ Teams handling regulated or unusually confidential conversations should therefore evaluate the DPA, BAA where applicable, retention/deletion controls and configuration of connected accounts rather than treating CRM ingestion as ordinary low-sensitivity SaaS data.

Lightfield explicitly relies on third-party services, including AI/ML providers, and its Terms warn that AI output can be inaccurate, incomplete or hallucinatory and can be subject to provider terms.¹⁰ This makes review controls, permissions and source citations part of the product’s trust model rather than optional user discipline.

Lightfield is best aligned with startups and forward-deployed B2B teams whose customer reality lives in calls, inboxes and meetings but whose CRM falls behind because nobody wants to maintain it. Founders, account executives, RevOps, customer success and product teams can all benefit when the same customer memory is queryable instead of being scattered across notes, meeting tools and stale CRM fields.¹¹¹

It is especially interesting for teams willing to reorganize their GTM stack around an agentic CRM rather than add another AI assistant to an existing one. The combination of automatic capture, flexible memory, automations, outbound sequences, APIs/MCP and artifact generation can reduce handoffs between multiple point solutions.¹⁶⁷

The fit is weaker for organizations that primarily need a cheap, stable system of record with predictable per-seat costs. Lightfield’s own usage guidance implies four-figure monthly economics per heavily active seller, and the public billing documentation currently conflicts enough that procurement should verify the exact commercial model.⁵⁹¹⁰

Large enterprises with extensive legacy-CRM customization should also validate coexistence and migration carefully. The product can expose a powerful API, but Lightfield itself says direct CRM integrations are not yet the normal native path.⁶ Teams that require a long operating history, large verified-review populations or independently measured reliability should treat Lightfield as a promising but still lightly validated platform and run a controlled pilot on representative customer data.