Macha in 2026 So Far: A Feature-by-Feature Recap

This is a straight, dated recap of every meaningful shipment from the Macha changelog between February 22 and June 23, 2026. No roadmap promises, no marketing gloss on things that don't exist yet — just what actually landed, grouped so you can find the capability you care about. If you've been away for a quarter, start here.

A quick orientation before the timeline, because it frames everything below: Macha is an AI agent layer that sits on top of the helpdesk you already run — Zendesk, Freshdesk, and now Gorgias and Front. It doesn't replace any of them. It connects your commerce, knowledge, and comms tools, and builds agents that resolve, triage, and route tickets while taking real actions through those tools. Almost every release this year pushes on one of three axes: more ways to build agents, more places to connect them, or more control over what they cost and do.

The big picture: 2026 in five moves

The sections below walk each theme in the order things shipped.

Building agents: from a form to a conversation

The single biggest shift this year is how you create an agent. In February it was a configuration form. By April it's a conversation.

Sub-agents (March 8, expanded April 9). Agents can call other agents as tools. You build a specialist — a refunds agent, a shipping-status agent — and a coordinator delegates to it. By the April expansion this goes up to three levels deep with circular-reference prevention, and sub-agents gained write permissions in autonomous mode (they stay read-only in interactive chat for safety). In June, sub-agent runs became fully auditable in chat: each delegation produces its own nested, expandable conversation card showing the sub-agent's model, tools, every tool call and result, and the final reply.

AI Agent Builder & Editor (April 3, hardened April 9). This is the headline. Describe what you need in plain language and Macha generates a fully configured agent — instructions, tools, triggers, sub-agent links, and model selection — and it's available on all plans, including Trial. The builder is context-aware: it knows which integrations you've connected and suggests relevant configurations. Created agents, triggers, and tools show up as clickable artifact cards in the chat, so you can jump straight from the conversation into the thing it built. The same AI editor opens from any existing agent to change it by talking to it.

AI Tool Builder & Custom API Tools (March 30 → April 9). Agents stopped being limited to Macha's built-in connectors. Custom API Tools let you wire any REST endpoint — URL params, body templates, API-key / bearer / basic auth, credentials encrypted at rest. The AI Tool Builder does it conversationally: describe the API, paste credentials, and Macha discovers endpoints, tests them, and creates the tools, auto-marking POST/PUT/PATCH/DELETE as write operations that require confirmation. Tools group under named connectors with their own icons and brand colors.

If you only adopt one thing from the first half of 2026, the Agent Builder is it — it collapses the build step from "learn the form" to "say what you want."

Connecting more tools: from two helpdesks to four

Macha's reach expanded steadily, and the connector list is the clearest signal of where it's heading.

• Freshdesk (March 15) — 15 tools covering the full ticket lifecycle (get/search, replies and notes, priority/status/tags/fields, assignment, contact search, KB articles), plus attachment text extraction from PDF, DOCX, XLSX, CSV, and TXT, and webhook triggers.
• Shopify (March 22) — OAuth connection with 7 tools: search products, look up and search orders, find customers, view discounts, and process refunds (with confirmation). This is the commerce layer that makes an order-status or refund agent real.
• Confluence (April 9) — 7 tools (search, read, list spaces, child pages, create, update, comment), with token-expiry monitoring that emails you 30 days before your Atlassian token lapses. Connecting it auto-creates a knowledge source, the same way Google and Notion do.
• Gorgias and Front (June 23) — two new helpdesk connectors, 15 tools each. Gorgias targets Shopify-native support teams; Front is shared-inbox-first with @mention support on internal comments. Both connect via API key under Connectors → Support with starter agent templates.

Underpinning all of this, April 9's Zendesk OAuth milestone matters: Zendesk approved Macha's global OAuth client, making OAuth the default for new Zendesk connections (replacing API-key auth) with automatic token refresh. And March 8's multi-instance connectors let you link several accounts of the same type — two Zendesk instances, three Stripe accounts — with automatic tool disambiguation so the agent knows which one to use.

See the full list on the integrations page.

Running agents autonomously: triggers, schedules, and guardrails

An agent that only answers when you open a chat is a toy. The trigger system is what makes Macha operate on its own.

Trigger system (March 1). Agents run autonomously in response to external events — Zendesk ticket events, Freshdesk updates, Slack mentions, and custom webhooks — with configurable debounce to batch rapid events before firing.

Scheduled triggers (March 22, detailed April 9). Cron-based execution for agents that need to run on a clock — a daily backlog sweep, an hourly SLA check. Sensible guardrails ship with it: minimum 1-hour intervals, a 24-runs/day cap, org concurrency limits, and auto-disable after 5 consecutive failures with an email. Pro plans get 3 scheduled triggers; Enterprise gets 20.

Custom webhooks (March 1 / April 9). A universal webhook URL with a signing secret that works with any system — no connector required. Test Run (April 9) closed the confidence gap: pull a real ticket, email, or document and watch the agent process it live before you turn it loose, with a simulation schema defined across all 14 connectors at the time.

Per-event debounce (June 5). Debounce used to be messaging-only. Now any event-driven trigger — Zendesk/Freshdesk ticket events, Slack mentions or DMs — exposes a debounce chip, with safeguards that force-fire a bucket at 25 accumulated events or a 2-minute cap so a malformed event stream can't hold an agent hostage.

A subtle but important June 5 fix: toggling an agent off no longer deletes its Zendesk Triggers — it deactivates them, preserving the resource IDs so any Zendesk Trigger you built in your own Admin Center keeps working across pause/resume cycles.

Models and vision: a cleaner, cheaper lineup

The model story in 2026 is "more capable, then cheaper."

Image vision on tickets (April 17). When an agent hits an image attachment on a Zendesk ticket (JPEG, PNG, GIF, WebP), it downloads, encodes, and sends it to a vision-capable model — so it can read a screenshot, parse a receipt, or identify a product instead of asking the customer to describe it.

The new model lineup (April 23). Macha consolidated onto OpenAI's GPT-5 family: GPT-5.4 (OpenAI's most capable model, with a 1M-token context window per OpenAI), GPT-5 as the new default (also 1M context), and GPT-5.4 Mini as the fast, cheap workhorse. Three older models — GPT-5 Mini, GPT-4o Mini, and o4-mini — were retired, with existing agents auto-migrated to GPT-5.4 Mini so nothing broke.

Price cuts (June 17). Two months later, the credits came down: GPT-5 dropped to 2 credits per response (from 3) and GPT-5.4 to 3 (from 5), while GPT-5.4 Mini stayed at 1 — the default for most workflows. Same plan, same allowance, roughly two-thirds more premium-model responses. Worth restating the model here: credits are charged per AI action, deducted once per complete response (0.5–9 credits depending on model, with GPT-5.4 Mini at 1) — not per deflection or resolution. See pricing for the live numbers.

Studies: batch AI analysis of your whole queue

Studies (May 30) is the year's standout new surface area — not an agent, but an analysis tool. Point it at a batch of records (Zendesk tickets to start), define typed columns the AI fills in (Yes/No, choice, number, text), and it processes the whole batch in parallel into a structured grid. Critically, it shows you the exact record count, per-record credit cost, and total before any run, charges only for records that successfully process, caps runs at 20,000 records, and can push the finished analysis into a Knowledge Source your agents then search.

The Report view (June 5) turned the grid into insight: stat callouts, bar charts for Yes/No and choice columns with counts and percentages, and histograms for number columns with min / mean / median / max — click any bar to drill into the matching records.

Studies is on Professional and Enterprise; the Report view is Professional and Enterprise too. We did a full first-party walkthrough — a real 57-ticket run with the numbers — in our Studies deep-dive.

Billing: top-up packs and honest cost controls

Spending got more flexible and more transparent in 2026.

Credit top-up packs (May 30). Three one-time packs on top of your monthly allowance — 1,000 credits for $99, 5,000 credits for $449, and 10,000 credits for $799 — that never expire and are spent automatically after your monthly credits run out. Available to active Starter and Professional subscriptions, with idempotent webhook handling so a duplicate delivery can't double-credit you.

That sits on top of the credit system foundation from February 22 (per-model pricing, deducted once per complete response, usage alerts at 50/80/90%, in-app Chargebee checkout) and April 23's pricing reset ($299/mo for Starter, $699/mo for Professional). A reliability fix on June 17 also made subscriptions recover automatically after a failed-then-retried card payment, instead of getting stuck in a past-due state.

The quality-of-life layer (don't skip this)

Not every shipment is a headline, but the small stuff is where daily friction lives:

• Soft-delete agents (June 17) — deleting an agent now moves it to a 30-day Trash with one-click restore; triggers are paused (not destroyed) so a restore brings everything back intact.
• Onboarding & welcome chat (March 8 → June 23) — a 3-step guided setup, capped by a no-credit welcome conversation that uses your connected tools to surface real recent tickets/payments and suggests next steps.
• Connector health (April 23) — connectors track a connected/auth-failed status, auto-deactivate affected agents on auth failure and email admins, with one-click test-connection and in-place reconnect flows.
• Agents-list columns (June 10) — model, tools, triggers, and sub-agent counts now scan at a glance from the table.
• GDPR & data controls (April 9) — 30-day scheduled account deletion, JSON data export (with secrets stripped), cookie consent, a 45-day conversation TTL, and AES-256-GCM encryption of credentials in logs.

Watch-outs: what this recap doesn't promise

A few honest caveats so you read the timeline correctly:

• Some features are plan-gated. Studies and its Report view are Professional/Enterprise; scheduled-trigger counts differ by plan. The Agent Builder, encouragingly, is on every plan including Trial.
• One genuinely "coming soon" item. The Ask Sidekick bar on the Studies Report (June 5) ships as a deliberately disabled preview — study-scoped Q&A isn't live yet, and Macha keeps it inert rather than risk a misleading answer. That's the only roadmap teaser in this whole recap; everything else above is shipped.
• Connectors are helpdesk-and-tools, not everything. The native helpdesk connectors are Zendesk, Freshdesk, Gorgias, and Front. Anything else you need, you reach through Custom API Tools — powerful, but it's a build step, not a one-click connector.
• This is "so far." It covers Feb–Jun 2026. Check the changelog for anything that landed after this post's date.

For context on why this pace matters: Gartner projects AI-agent software spending will hit roughly $206.5 billion in 2026, up 139% year over year — agents are moving from pilots to budget lines, and the platforms shipping fastest are the ones worth betting on.

FAQ

What's the single biggest Macha feature of 2026 so far? The AI Agent Builder (April 3), which lets you create a fully configured agent — instructions, tools, triggers, sub-agents, model — from a plain-language description, on every plan including Trial.

Which helpdesks does Macha connect to now? Four native helpdesk connectors: Zendesk and Freshdesk (shipped earlier and through Q1), plus Gorgias and Front added June 23, 2026. Macha layers on top of these — it doesn't replace them.

Did pricing change in 2026? Yes — twice in spirit. April 23 reset plan pricing and the model lineup; June 17 cut per-response credit costs for GPT-5 and GPT-5.4 (GPT-5.4 Mini stayed at 1 credit). Credits are charged per AI action. See the pricing page for current numbers.

What is Studies and when did it ship? Studies (May 30, Report view June 5) runs an AI analysis across a batch of records — define typed columns, get a structured grid plus charts, and pay only for records that process. It's on Professional and Enterprise.

Where do I see the official, dated list? The Macha changelog is the source of truth and is updated continuously. This post is a curated recap of it through June 23, 2026.

The throughline

Read the dates end to end and a strategy shows up. The first quarter was about foundations — connectors, triggers, knowledge sources, billing. The second was about leverage: building agents by talking, delegating to sub-agents, analyzing the whole queue at once, and trimming what every action costs. Macha spent 2026 making it dramatically cheaper — in effort and in credits — to put a capable agent in front of real tickets.

If you've been on the fence, this is a good moment to look again. Start a 7-day free trial, no credit card required, connect the helpdesk you already use, and ask the Agent Builder to build your first agent — or read the docs for the full walkthrough.

Sources: OpenAI — Introducing GPT-5.4 · DataCamp — GPT-5.4 · Gartner AI-agent spend forecast (via Crescendo AI news roundup)

Written by Abbas (Customer Support & AI, Macha) · Reviewed by Ankeet Guha (Co-founder & CTO) · Published 2026-06-24 · Last updated 2026-06-24.