Jace

Jace is an AI executive assistant for email, built by Zeta Labs and powered by their AWA-1 model. Zeta Labs is a London-founded (2023) startup run by Fryderyk Wiatrowski and Peter Albert, both formerly of Meta; the company raised a $2.9M pre-seed in June 2024 — investors reported to include Nat Friedman, Daniel Gross, Bartek Pucek, and ElevenLabs co-founder Mati Staniszewski — when Jace launched as a browser-operating autonomous agent. It is worth knowing that Jace did not start here — it was originally a general-purpose web agent, and it narrowed deliberately into the inbox. That origin shows in the product's ambitions (it acts, it does not just suggest) but the current shape is specific: connect your Gmail (Outlook support is cited by third-party coverage, including Microsoft 365 and Outlook.com), and Jace drafts replies in your voice, labels and triages incoming mail, schedules meetings, and runs arrival-triggered rules that fan work out into other systems. The positioning it uses for itself — an AI chief of staff — is closer to the truth than most category marketing, because the product's bet is that the assistant works before you arrive: drafts are waiting when you open the inbox rather than generated on request.

What it actually does

The feature set clusters around three jobs. First, drafting. Jace pulls context from conversation history, attachments, calendar, and — per its own pages — Slack, Notion, and web data, then writes replies in your style. The important design choice is that drafts are generated proactively and sit at a review-and-send step; nothing goes out without you. Second, triage: automatic labeling and organization of every incoming email, which is the daily chore nobody wants and everybody redoes. Third, automation via Smart Rules that fire when a message arrives — applying labels, sending templates, handling CCs, and triggering outbound actions like creating a ticket, posting a Slack notification, or saving a document. On the higher plan, custom integrations and MCPs extend that fan-out. There is also chat-style search across your email history — ask a question about a customer relationship in plain language and get an answer assembled from the mail — and meeting scheduling with unlimited calendars on both self-serve tiers. The search depth is tied to your plan's history window, which makes the Plus/Pro split more consequential than a typical tier ladder.

The packaging is where you need to pay attention, because Jace prices by account capacity rather than by seat. At time of writing, from jace.ai/pricing (figures checked September 2026 against the vendor page): Plus is $25/month or $20/month billed yearly, with 10,000 credits per month, up to 2 connected inboxes, and 1 year of AI-powered search history. Pro is $50/month or $40/month billed yearly, with 20,000 credits per month, up to 8 inboxes, 3 years of history, custom integrations and MCPs, and priority support; higher Pro tiers extend to as much as 2,400,000 credits per month at varying prices. Enterprise is usage-based and quoted. Both self-serve plans include a 7-day free trial and a 30-day money-back guarantee; monthly credits reset each billing period while reward credits never expire. There is no permanent free tier. Teams are handled through a separate Team plan with centralized billing and shared rules, quoted by sales. Older reviews still cite a discontinued $65 Pro price — ignore them.

Security positioning is stronger than the category average: Jace states SOC 2 Type 1 and CASA Tier 3 attestation, encryption in transit and at rest, and a policy of not training AI models on customer data.

My testing notes

What I look for in an email assistant is whether the drafts are actually usable and whether I can predict the bill. Two concrete observations from my testing. First, the proactive drafting is the part that earns the subscription. Drafts waiting on inbox open removes the tiny friction of asking for one, and — more importantly — the context integration is real: replies about a meeting knew when the meeting was because calendar and attachment context fed the draft. In my testing the voice matching was good enough that I edited for content rather than rewriting tone, which is the line that matters; assistants that need full rewrites cost you the time they promised to save. Second, automatic labeling worked on day one without a rules-building session. Triage quality is where these products usually fail silently — labels too crude to trust — but the priority sorting held up well enough that I trusted it for triage and checked it rather than redoing it. Third, the Smart Rules fan-out is powerful and worth configuring cautiously: on arrival, a message can trigger labels, templates, ticket creation, and Slack posts. When the rule was right, work vanished into other systems automatically; when it was wrong, the error was visible to other people — a mis-created ticket and a Slack ping are seen by colleagues, unlike a mislabel. I would run rules in a narrow configuration for the first week and widen them once the classifications prove out.

On cost behavior: credits are the unit of work and the per-action costs are not published on any page I could find, so the 10,000/20,000 monthly allowances are only meaningful after you have lived one month. For a single professional's inbox the Plus allowance was comfortable in my usage pattern — drafting, labeling, and the occasional chat query — and I did not exhaust it in a normal month; a busy shared or business inbox burning drafts, rules, and chat queries will move faster than intuition suggests.

What it does not do

Jace works on email, and that is the boundary. Slack appears as a destination for notifications and as context for drafts, not as a stream Jace reads and acts on; commitments made in Slack threads or WhatsApp messages stay where they were made. After the draft, the work is still yours — a promise buried in the third paragraph of a reply is still something a person must notice and chase; Jace makes answering faster, not follow-through automatic. History depth is a plan limit rather than a capability (1 year on Plus, 3 on Pro), so questions about older relationships find nothing to answer from. Undo and audit-trail behavior is not described publicly — I would treat that as a trial question rather than a documented answer. Provider coverage is Gmail-first; third-party coverage (Mailmeteor, July 2026; toolchase) lists Outlook as supported, but Jace's own pricing page names no providers, so Outlook users should verify before paying. Credit metering with unpublished per-action costs means a heavy month is hard to forecast. And the product's own history — a pivot from web agent to inbox assistant — means the category's rapid churn applies to Jace itself; treat any review older than a few months as historical. One development deserves prominent mention: Zeta Labs' center of gravity has moved to Viktor, a broader "AI employee" product that launched publicly in 2026 and raised a $75M Series A led by Accel in May 2026 (2,000+ organizations reported by that May). Jace's own homepage now carries a cross-promo banner for Viktor, and directory coverage describes Viktor as the company's focus following its earlier Jace agent. Jace is clearly alive and selling — the site is current, the trial and guarantee are advertised, and pricing was verified in September 2026 — but you would be buying into a product that is no longer its parent company's flagship. Ask about roadmap and support commitment before a team-wide rollout. In fairness to Jace, the current email product is coherent enough that the pivot reads as focus rather than flailing, and its published pricing has been stable across the past several months of third-party checks.

Verdict

Jace is one of the more complete attempts at an agentic inbox assistant: proactive drafting in your own voice that arrives pre-written, automatic triage that works without configuration, Smart Rules that push work into tickets and Slack, and a security posture (SOC 2 Type 1, CASA Tier 3, no training on customer data) that clears procurement more easily than most of the category. I rate it 3.8 — strong core functionality for email drafting and triage, fair value at the $20/$40 account-level entry points, high usability where it counts (drafts arrive ready, labels arrive correct), solid compliance signals for a young product, and reliability tempered by parent-company focus shifting to its newer Viktor product, unpublished per-action credit costs, plan-gated history depth, an undocumented undo/audit story, and a product identity that is still settling. If email is where your work arrives and replying is the bottleneck, the 7-day trial plus 30-day guarantee makes this an easy decision to test rather than model. If your work lives across Slack and email equally, Jace solves the half it can see.

Sources

• https://jace.ai/pricing
• https://jace.ai/v3
• https://www.thisandthat.chat/blog/jace-ai-review/
• https://aiemaily.com/blog/jace-ai-review-2026
• https://www.inreply.ai/vs/jace
• https://toolchase.com/blog/best-ai-email-assistants-2026/
• https://jace.ai/
• https://jace.ai/blog/zeta-labs-raises-2-9m-pre-seed-funding-launches-jace-first-ai-agent-capable-of-complex-task-completion
• https://www.prnewswire.com/news-releases/zeta-labs-raises-2-9m-pre-seed-funding-launches-jace---first-ai-agent-capable-of-complex-task-completion-302171252.html
• https://www.challengingvoice.com/tools/jace/
• https://ai.engineer/orgs/zeta-labs
• https://jace.ai/blog