Executive Product Overview

Most project-driven organizations do not have one problem with their tooling. They have four systems that each solve a quarter of it:

• Work management lives in a board or tracker.
• Client communication lives in email and shared documents.
• Delivery tracking lives in a spreadsheet, or in an engineering tool nobody outside engineering opens.
• Reporting lives in a deck that somebody rebuilds by hand every week.

None of those systems is wrong on its own. The cost is in the seams. Somebody has to copy the status from the tracker into the client update. Somebody has to reconcile the timesheet against the invoice. Somebody has to notice that a task has been sitting in review for nine days. That work is invisible on any roadmap, it is done by the most senior people available, and it scales linearly with headcount.

Adding another point tool adds another seam. The alternative most teams reach for — an AI chat assistant bolted onto the side of the existing stack — can describe the glue work, but it cannot do it, because it has no permission to write into any of those systems and no accountability if it does.

Kavanah is one workspace where the work, the client relationship, the delivery signal, and the reporting all live in the same data model — and where an AI agent operates on that data model directly, under the same permissions, audit trail, and approval rules as a human member of the team.

Concretely, a Kavanah workspace holds projects, tasks and sub-tasks, clients, discussions and chat, time entries, documents, planning horizons, requests from outside the company, and connected systems (mail, calendar, code, finance). The AI agent reaches all of it through roughly 896 registered tools — not a chat window that returns text, but an agent that creates the task, drafts and sends the update, reassigns the work, and records what it did.

Because every action — human or agent — passes through the same API, the same tenant boundary and the same audit log, the governance story does not change when you turn the AI on. See the Architecture Overview and Product Data Sheet for the structural detail.

Kavanah suits organizations that run identifiable projects for identifiable stakeholders. It is a weaker fit for pure always-on-operations teams with no project structure, and there is no self-hosted or on-premise deployment today — Kavanah is delivered as multi-tenant SaaS with production data processed in the United States.

The full functional inventory is in the Capabilities Matrix and the Product Data Sheet.

AI that acts, under governance

The common pattern in this category is a chat sidebar: it can read your data and suggest text. Kavanah's agent is a governed tool-calling loop over the product's own API. It runs on Anthropic Claude models and reaches roughly 896 registered tools, each of which enforces the same permission and workspace-scoping checks as the user interface.

The controls around that are the point, not an afterthought. A safe mode strips write tools entirely. An action-policy engine can park risky actions for human approval before they execute. An action ledger records what the agent did, and an undo stackreverses it. Custom agent identities — what we call AI Employees— carry their own instructions, model, capability scopes, and member access control list, so “which AI can do what, for whom” is an administered decision rather than a prompt. Our data-use posture is documented on the AI Transparency page.

One codebase, every platform

Web, desktop (macOS, Windows, Linux) and mobile (iOS, Android with native push) are all shells over the same deployed application. Features and security fixes reach every platform at once; there is no stale mobile client running last quarter's permission model, and no separate mobile roadmap to wait on.

Clients are first-class, not an add-on

Client records and scoped client portals are part of the core data model. External stakeholders get visibility into their own work without being given a seat inside the workspace and without a parallel status document that drifts.

Open by API, not a walled garden

Everything the interface does, the API does: a documented HTTP API with an OpenAPI 3.1 specification covering 500+ endpoints, published at /openapi.yaml with a reference at /docs. Workspace API keys inherit the permissions of the user who minted them, so integration access cannot exceed human access.

Priced per seat, without the enterprise tax

Paid plans start at $3 per seat per month with uncapped seats. AI usage is metered in credits with an allowance included in every paid seat, so the cost of automation is visible and controllable rather than bundled into an opaque platform fee. See the pricing sheet.

If you need to establish these outcomes for your own organization before committing, the proof-of-concept process exists to measure them against your success criteria rather than ours.

Kavanah runs on Vercel with a managed Neon Postgres database, with production data processed in the United States. Data is encrypted in transit (TLS 1.2+ with HSTS) and at rest (AES-256), with an additional application-layer AES-256-GCM envelope over integration credentials, OAuth tokens, MFA secrets, and classified PII — keyed to per-workspace subkeys so no two tenants share an encryption key. Tenant isolation is enforced on every workspace-scoped table and checked by a continuous-integration linter that fails the build if a query is missing its tenant filter.

Identity supports SAML 2.0 and OIDC single sign-on, TOTP multi-factor authentication, OAuth sign-in, and SCIM directory sync. Audit logs are immutable, enforced by database triggers, with configurable retention (7+ years), legal hold, and eDiscovery export. GDPR and CCPA data-subject export and erasure are supported.

Certification status, stated plainly

• SOC 2 Type II— in its observation period, not yet certified.
• ISO/IEC 27001— planned, targeting Q4 2027.
• HIPAA BAA— available on Enterprise.
• Penetration testing — annual; summary available under NDA.

Full detail sits in the Security Overview, Technical Data Sheet, DPA, Sub-processors, Business Continuity, and the SLA. Security and procurement questions go to security@kavanah.ai.

Kavanah is priced per seat, with AI usage metered separately in credits. Seats are uncapped on every paid plan — the plan sets the project, storage, and AI-credit allowances, not the size of your team.

• AI credits beyond the included allowance are purchased separately as credit packs, at roughly $0.01 per credit.
• Yearly billing is arranged through sales rather than self-serve checkout.
• Enterprise adds custom commercial terms; a HIPAA BAA is available on that edition.

Line-item detail, the credit model, and what a procurement team needs for a purchase order are in the Pricing Sheet. For a quote or a yearly agreement, use the contact form.

A typical enterprise evaluation runs in four stages. Each has a document behind it so your team is not waiting on a call to make progress.