Intercom Alternative for AI Support — No Per-Resolution Fee

An open-source Intercom alternative for AI support should target the Fin automation layer, not pretend to replace the entire helpdesk. It can remove the vendor's per-outcome meter, but it does not remove cost or reproduce Fin's Apex model, channels, analytics, escalation UX, and support operation. The lowest-risk path is to keep your current helpdesk, use Eigent for one bounded retrieval-and-drafting workflow, require human approval, and compare total cost per successful outcome before expanding.

First, the Intercom and Fin naming update

The company formerly called Intercom renamed itself Fin on May 12, 2026, while the helpdesk product kept the Intercom name (Fin announcement). “Intercom Fin alternative” remains useful search language, but the current company identity is Fin.

Salesforce signed a definitive agreement on June 15, 2026 to acquire Fin for approximately $3.6 billion, subject to adjustments and closing conditions (Salesforce press release). As of the research date, the accurate wording is “agreed to acquire,” not “owns.” The announced deal creates legitimate roadmap and procurement questions, but it is not evidence that pricing or the product will change.

What Fin costs at volume

Fin's current chat and email schedule charges $0.99 for a resolution, configured Procedure handoff, or disqualification, and $9.99 for a sales qualification. Voice, high-volume, and specialized arrangements are sales-led (Fin outcome documentation). Buyers can configure reminders and hard limits.

These are arithmetic examples, not quotes. They assume every listed item is billable at $0.99 and exclude seats, optional channels, implementation, and negotiated high-volume terms.

Intercom's annual-billing helpdesk prices are currently $29 per full seat each month for Essential, $85 for Advanced, and $132 for Expert (Intercom pricing). Twenty Advanced seats add $1,700 per month or $20,400 per year before outcomes and optional channels. Fin can also run on another helpdesk without Fin seat fees, although a minimum outcome commitment applies (Intercom pricing).

What counts as a Fin outcome?

The $0.99 headline needs its billing definitions.

• A resolution can be explicitly confirmed or assumed when the user leaves without requesting more help.
• If the user returns to the same conversation needing more assistance, the outcome is deducted.
• An abandoned clarifying question is not billable.
• A configured Procedure handoff can be billable.
• A default escalation is not billable.

These rules come from Fin's current outcome documentation (Fin help). Procurement should replay real conversations and audit assumed resolutions, reopen windows, configured handoffs, duplicates, retries, and channel transfers before accepting an annual forecast.

Fin says its Apex model handles essentially all English chat and email conversations, resolves almost two million issues per week, and had grown to nearly $100 million in recurring revenue by March 2026 (Fin Apex announcement). These are first-party product and business claims, not an independent benchmark.

Why teams look for an Intercom or Fin alternative

Outcome cost grows with successful automation

At high volume, a fixed price per billable outcome becomes a material operating line. In a March 2026 discussion, a commenter affiliated with competitor My AskAI argued that 5,000 resolutions creates roughly a $5,000 monthly AI layer and questioned whether assumed resolution always equals customer success (Reddit discussion). The affiliation and anecdotal nature matter; Fin's own billing definitions are the stronger source.

Source and deployment control

Fin is a proprietary managed platform. Teams that need to inspect the orchestration layer, run it locally, choose their model, or retain indexes and tool credentials inside their environment may prefer an open stack.

Ownership change

The rename and pending Salesforce acquisition make portability worth testing. Exporting prompts, policies, knowledge, transcripts, action schemas, and evaluation cases reduces dependency regardless of whether the deal changes the product.

Intercom and Fin vs an Eigent-based stack

Eigent's strongest role is a controlled orchestration layer. It can fetch a ticket, retrieve approved documentation, read CRM or order context through a scoped API, draft a response, classify risk, wait for human approval, and write the evidence and disposition back.

It does not provide Fin's helpdesk, proprietary support model, packaged omnichannel connectors, analytics, support-specific evaluation system, mature escalation interface, contractual support, or reported scale.

A safe reference workflow

1. Receive one low-risk ticket class. Start with an internal FAQ or status question, not refunds, identity changes, or regulated advice.
2. Classify intent and risk. Route sensitive, ambiguous, angry, or account-changing requests directly to a human.
3. Retrieve approved sources. Use versioned policy and knowledge content; attach citations to the draft.
4. Read account context with least privilege. Prefer a narrow API over a general browser session.
5. Draft, do not send. The agent should propose the answer and any action separately.
6. Require approval. A support owner reviews source accuracy, tone, account data, and action.
7. Write back to the helpdesk. Preserve the response, sources, disposition, tool calls, and reviewer decision.
8. Evaluate reopens and corrections. A cheap answer that creates a second contact is not a successful outcome.

The customer-support domain workflow is a relevant starting point, but your existing helpdesk should remain the system of record during the pilot.

How to model total cost

Compare Fin's outcome and seat invoice with the open stack's full operating cost:

• model API or local GPU cost;
• helpdesk and connector fees;
• retrieval/index infrastructure;
• engineering and integration time;
• evaluation-data maintenance;
• security review and incident response;
• human review time;
• support and availability coverage.

Use cost per successful outcome, defined by policy adherence, no avoidable reopen, correct action, acceptable latency, and a satisfied or appropriately escalated customer. Do not optimize only for containment.

Migration path without replacing the helpdesk

Shadow mode

Run the open workflow against historical or live copied tickets without sending responses. Compare its drafts with human resolutions.

Human-approved drafts

Allow agents to prepare answers inside the existing queue. Require a representative to review and send every message.

Bounded actions

Add one reversible, low-risk action such as retrieving an order status. Keep refunds, cancellations, identity changes, and financial adjustments human-controlled.

Controlled automation

Automate only the ticket classes that meet a sustained quality threshold. Retain event-level logs, spend limits, rollback, and a route to human support.

For wider category choices, see /blog/open-source-ai-customer-support-agents.

Open building blocks for an Intercom alternative

Chatwoot's Community code is MIT-licensed while its enterprise/ directory is separately licensed; paid self-hosted plans include additional support and AI features (Chatwoot license, Chatwoot pricing). Zammad is an AGPL-3.0 self-hostable helpdesk (Zammad repository). The legacy Rasa framework repository is Apache-2.0, while current Rasa Pro/CALM is license-key software (Rasa repository, Rasa licensing).

These components cover different layers. A helpdesk is not an AI agent, a conversation engine is not a ticketing system, and an orchestration workspace is not a contact center.

When should you keep Fin?

Keep Fin when the support-specific model, omnichannel product, managed evaluation, analytics, implementation, and enterprise service justify outcome pricing. Consider an open stack when source and deployment control are mandatory, the workflow is narrow, and your team can operate the missing helpdesk, connector, security, evaluation, and support layers.

Remove the meter carefully

Eigent removes a per-resolution application fee; it does not remove the cost or responsibility of reliable support automation. Start with one approved workflow through the customer-support solution, retain your helpdesk, and measure reopens and corrections. Download Eigent to run the shadow-mode pilot.