The best AI voice agents for logistics, judged on open source, self-hosting and on-prem, cost and logistics fit, are Dograh, HappyRobot, Retell AI, Vapi, Bland AI, Synthflow and Telnyx. Dograh is the only one that meets all four criteria. HappyRobot leads at the largest brokers with no public price. Every figure here comes from the vendor's own pricing page.
Key Takeaways
- Only one platform in the set is open source and free to self-host.
- Two vendors quote $0.05 a minute, and the two numbers buy different things.
- The top freight brokers buy a closed product with no public price.
This post is part of our guide to AI voice agents for logistics and delivery. The guide's section on couriers, freight and the build-or-buy choice points here for the buy side: seven platforms compared on the same four criteria, and the gap below the enterprise tier. Dograh, an open-source, self-hostable voice agent platform, is one of the seven, judged on the same criteria as the rest.
The four criteria we used to rank the platforms
Four criteria, applied the same way to every product, decide the order below. Open source: can you read the platform code and run it without asking? Self-hosting and on-prem: can the platform run on servers you control, and at what price? Cost: what does a minute cost at list price, and what does that price include? Logistics fit: does the product have the features a delivery or freight call needs, such as reading a tracking system during the call, starting a call from a shipment event and writing the result back?
The set is voice-first platforms only, meaning products that place and answer phone calls as their main job. Chat-first customer service suites with a voice channel bolted on, and supply chain planning software with "agent" in the name, are different purchases and are left out.
How the seven platforms compare on each criterion
Dograh is first because it is the only product that meets all four criteria. The rest are ordered by how much of a logistics operation they can serve without a sales conversation.
| Platform | Open source | Self-host or on-prem | List price | What you get, and what's missing |
|---|---|---|---|---|
| Dograh | Yes, BSD 2-Clause | Self-host the full platform for free; Dograh Cloud as the managed option | No per-minute platform fee self-hosted (servers and paid model keys still cost); Dograh Cloud: 1¢/min platform fee | Visual workflow builder, Pre-Call Data Fetch, HTTP API tool with tool Messages, webhook node, API Trigger, any model by key or self-hosted. But community support on the free version, and call transfer to a person works on Twilio only |
| HappyRobot | No | Vendor cloud; no deployment statement | No public pricing | Agents across eight industries, best known in logistics for carrier sales and track and trace, in production at the largest US brokers. But closed, sold by demo, and with no price anyone outside a sales call can see |
| Retell AI | No | Vendor cloud on AWS; no services in the EU | $0.055/min voice infra; $0.07-0.31/min all-in by its own range | Free SIP trunking, BAA at no additional fee, SOC 2, a logistics page promising load booking and tracking calls. But closed, the TMS names appear only in its FAQ, and models are billed per minute on top |
| Vapi | No | Vendor cloud; no on-prem or VPC statement | $0.05/min hosting fee, before models and phone | Large model and voice marketplace, a forkable last-mile delivery template. But HIPAA handling is a $2,000 a month add-on, and the hosting fee excludes every model |
| Bland AI | No | On-prem or VPC on Enterprise only | Start $0.14/min; Build $299/month plus $0.12/min | Per-minute rate includes the model, speech-to-text and text-to-speech; BAA on Enterprise. But 100 calls a day on Start, and the on-prem option is closed code behind a custom contract |
| Synthflow | No | EU or US hosting badges; no self-host statement | Enterprise contracts from $30,000 a year | Enterprise onboarding, native telephony or SIP. But one plan, no published per-minute price and no minutes figure |
| Telnyx Voice AI | No | Vendor cloud; no self-host statement | $0.05/min voice engine bundling speech-to-text and text-to-speech; LLM extra | Carrier-built, telephony on the same network, Growth from $2,000 a month. But 60-second rounding on every call, and the model is billed separately |
Dograh: the open-source platform you can run yourself
Dograh is the only entry whose code you can read before you sign anything. It is released under the BSD 2-Clause licence and has more than 5,800 stars on GitHub. Dograh is the orchestration layer: it runs the call flow, tools and telephony, and the speech and language models are a separate layer you choose. You can use Dograh's own models, bring your own key for any supported provider, or run open-weight models on your own servers.
On cost, self-hosting removes the platform line from the bill. There is no per-minute platform fee when you run Dograh yourself; you pay for your servers and for any paid model keys or phone minutes. Dograh Cloud, the managed version, charges a 1¢-a-minute platform fee. A self-hosted instance handles 10 concurrent calls by default, adjustable with a setting, and your servers and phone provider set the real ceiling.
On logistics fit, four features do the work. Pre-Call Data Fetch asks your system who is calling when the call connects, so the agent can greet a consignee by name. The HTTP API tool reads a tracking system or TMS during the call, with tool Messages that speak while the lookup runs. The webhook node writes the outcome back after the call. The API Trigger starts a call from a shipment event, passing the shipment details in the initial context. Campaigns dial a list from a CSV for driver sweeps or claim follow-ups.
What's missing is worth saying plainly. The free version comes with community support rather than a service level agreement; Enterprise adds support and custom SLAs. Transfer Call, the tool that hands a call to a person, works on Twilio only. Head-to-head comparisons with two of the hosted platforms are in Dograh vs Vapi and Dograh vs Retell AI.
Open Source Alternative to Vapi / Retell
Self-hosted voice agent platform — no per-minute fees
dograh-hq/dograh
Star on GitHub
HappyRobot: the incumbent at the top of freight
HappyRobot is the product the largest brokers already run. It sells "AI workers" across eight industries, and logistics, especially carrier sales and track and trace, is where it is strongest. It names DHL and Kuehne+Nagel as customers, and, according to Forkast, reports more than 150 customers with eight of the ten largest US freight brokers running its agents in production. Those are company-reported figures.
What's missing is a price. HappyRobot publishes no pricing; its site offers a demo booking and nothing else. The code is closed and the site makes no statement about where the platform runs. For a top-ten broker with a procurement team, that is normal. For the thousands of brokers, carriers and 3PLs below that tier, it means the most logistics-specific product on the market is one they cannot evaluate, price or run without a sales process. The case for everyone outside that tier is made in AI voice agents for freight and trucking.
Retell AI and Vapi: hosted platforms priced by the minute
Retell AI and Vapi are the two hosted platforms most logistics teams will shortlist first, and both bill the platform by the minute.
Retell AI. Retell charges a voice infra fee of $0.055 a minute and states its own all-in range at $0.07 to $0.31 a minute on pay-as-you-go, with the model and phone line billed on top; both figures are on Retell's pricing page. SIP trunking and custom telephony carry no charge. A business associate agreement (BAA) is available at no additional fee, and Enterprise lists a custom BAA. Retell holds SOC 2 Type 1 and Type 2 and runs on AWS, and it states that it does not currently operate services within the European Union.
Retell's logistics page promises load booking within company guidelines, rate negotiation and tracking calls to shippers and carriers. But the transportation management systems it names, DAT, Truckstop.com and McLeod, appear only in its FAQ, while its integrations section shows CRM and contact centre logos. Twenty concurrent calls are included and each extra one costs $8 a month. The full breakdown is in Retell AI Pricing Explained.
Vapi. Vapi charges what it calls a hosting fee of $0.05 a minute, listed on Vapi's pricing page before any model, voice or phone cost. HIPAA-eligible data handling with a BAA is a $2,000 a month add-on. Concurrency is 4 calls with no package, 10 on Core and 30 on Pro. Telephony is charged by the provider, not Vapi. Vapi offers a forkable last-mile delivery template that checks status by tracking number and reschedules or cancels, which is a fair starting point, though it names no logistics integrations.
But Vapi's page makes no on-prem or VPC statement, and the hosting fee buys the platform alone; every model minute is extra. The arithmetic is in Vapi Pricing Explained.
Bland, Synthflow and Telnyx: what each one sells
The remaining three sell the same job three different ways.
Bland AI. Bland sells plans. Start costs $0.14 a minute with no monthly fee, 10 concurrent calls and 100 calls a day. Build costs $299 a month plus $0.12 a minute, with 50 concurrent calls and 2,000 calls a day. The per-minute rate includes the model, speech-to-text and premium text-to-speech, which makes it the simplest price in the set to reason about, and the figures are on Bland's pricing page. Enterprise is custom and lists on-prem or VPC deployment, a BAA, SSO and data residency. But the daily call caps on the self-serve plans are low for a delivery operation, transfer minutes carry their own charge, and Bland's on-prem is closed code delivered through a custom contract.
Synthflow. Synthflow has one plan. Its pricing page says Enterprise contracts start at $30,000 a year, scoped by volume, concurrency, telephony, integrations and security, with EU and US hosting badges; that is the whole of Synthflow's pricing page. But there is no published per-minute price, no minutes figure and no self-host or on-prem statement, so a mid-size 3PL cannot estimate a bill without a call.
Telnyx. Telnyx is a carrier that added a voice AI engine. Its voice engine costs $0.05 a minute and that price covers orchestration, speech-to-text and text-to-speech; language model tokens and telephony are billed separately, per Telnyx's voice AI pricing page. Pay as you go has no minimum, Growth carries a $2,000 monthly minimum and Enterprise $5,000. Having the phone network and the agent on one bill is a real convenience, and Dograh runs on Telnyx numbers too, so the carrier is not a choice you have to make against the platform. But the voice engine applies 60-second rounding, charging each call up to the next whole minute, so short confirmation calls cost more than the per-minute figure suggests, and the page makes no self-host or on-prem statement.
Join the Dograh Community
Dograh is an OSS alternative to Vapi. Join our Slack community for queries, releases, best practices & community interactions.
What the two "$0.05 a minute" prices include
Vapi and Telnyx both quote $0.05 a minute, and the two numbers buy different things. Vapi's hosting fee covers the platform only; speech-to-text, the language model, text-to-speech and the phone line are all extra. Telnyx's voice engine fee already includes speech-to-text and text-to-speech, leaving the language model and the phone line extra. Retell's $0.055 sits between them, since Retell lists no separate speech-to-text charge.
Self-hosted Dograh has no platform line at all. You pay your servers, your model keys if you use paid providers, and your phone provider. The honest comparison is the total cost of a minute on your own call mix, which means adding the model and phone lines to each platform fee before comparing. A platform fee of five cents is the start of the arithmetic, never the end of it.
Picking the right one for your operation
The right choice depends on who you are. A top-ten broker with a procurement team and a seven-figure budget will get a demo of HappyRobot, and it will probably be good. A mid-size 3PL that wants to start this quarter will shortlist Retell, Vapi or Bland and should put the model and phone lines into the spreadsheet before picking on the platform fee.
A broker or carrier whose calls carry rate conversations, gate codes and driver phone numbers has a different question: where does that audio go? Only one product in the set lets the answer be "our servers", and only when you run both the platform and the models yourself, because bringing your own key to a hosted model still sends the audio to that provider. If a customer's security review or a contract asks that question, Dograh is the platform you can show them the code for. If the review never comes, pick on total cost per minute and the integrations you already have, and keep the agent's tracking lookups and write-backs in your own systems whichever platform you choose.
Glossary
- platform fee
- The per-minute charge for running the agent itself, billed before the speech models, the language model and the phone line. Self-hosted Dograh has none.
- voice infra fee
- Retell's name for its per-minute platform line, which also covers speech-to-text because Retell lists no separate charge for it.
- 60-second rounding
- Billing each call up to the next whole minute, so a 70-second confirmation call is charged as two minutes.
- self-hosting
- Running the platform's code on servers you control rather than in a vendor's cloud. Of the seven platforms here, only Dograh can be self-hosted without an Enterprise contract.
