OpenSEO, Opengeni, and AUDR: What the Free Version Still Costs You
Three of this week's launches have the same word in the pitch. Free. Open source. And every one of them still hands you a bill.
0:00 / 9:37
OpenSEO, Opengeni, and AUDR: What the Free Version Still Costs You
Three of this week's launches are open source, and every one of them still leaves you paying for something. The licence is the free part. What each one moves onto your team is the bill: a data vendor's invoice, a platform team's time, or an evening spent metering every layer of an agent run. OpenSEO 1 launched on October 8 and finished the week inside the top ten 2; Opengeni and AUDR reached the same board that week 1.
The short version. Choose OpenSEO when you want SEO data inside the coding agent you already pay for and DataForSEO-grade answers are good enough. Choose Opengeni when you are putting an agent inside your product and you don't already run Kubernetes. Choose AUDR when you are about to bill customers for agent usage and cannot yet say what one run cost.
| Tool | Best first job | Setup friction | Time to first value | Small-team price | Integration pain | Practical pick |
|---|---|---|---|---|---|---|
| OpenSEO | Get keyword, backlink and rank data into the agent you already use | One Docker container and your own DataForSEO key, or a hosted sign-up; the container starts with its own login disabled, so keep it off the open internet 3 | Minutes once the data key exists; the app has nothing to show without it | $10/month on the hosted plan, including $10 of usage; self-hosted has no licence fee, and you pay DataForSEO directly; hosted adds 28% to every data request 45 | It inherits another company's index, and it is pre-1.0 with most work done by one person 6 | A team that wants SEO data inside its coding agent and can live with DataForSEO-grade depth |
| Opengeni | Put a per-customer agent inside your own product | Cloud: sign in. Self-hosted: a Helm chart on Kubernetes, plus Postgres with pgvector, Temporal, NATS, object storage and a sandbox backend 7 | Cloud in minutes; self-hosted is a project, not an install | Model cost + 5% on the cloud, no seat fee or minimum; self-hosted is free software plus your infrastructure and model bills 8 | Outside managed mode a shared access key is, in their words, "a simple perimeter, not per-tenant access control" 7 | A product team that wants durable sessions, sandboxes and credentials solved without hiring for them |
| AUDR | Record what an agent run consumed, and whose it was | Install the SDK, wrap your runtime with an adapter, write records to a file; no account or pricing setup needed 9 | Minutes to a local record; the real work is getting a run id through every layer a run touches | The standard, the schema and the SDKs are Apache 2.0 and free; the spec carries no prices or rating logic | Five adapters and two sinks are published today, so a runtime outside that list means writing the adapter yourself 10 | A team that is about to bill for agent usage and needs the run-level record first |
OpenSEO: the code is free, the data isn't
OpenSEO is an open-source alternative to Semrush and Ahrefs: keyword research, backlinks, rank tracking, site audits, and a view of whether AI assistants mention your brand 11. The repository is MIT-licensed with more than 23,000 stars, and the product's declared direction is that a full SEO suite has become bloated and expensive, so your coding agent should ask for the data it needs over MCP instead 512.
The bill is the data. Every keyword, backlink and ranking comes from DataForSEO, a separate vendor you pay yourself; the open-source part is the software around that index 5. Hosted is one plan at $10/month including $10 of usage, and the hosted service charges 28% extra on every DataForSEO request it makes for you; top-up credits never expire, but an active subscription is required to spend them, and the free trial is $0.50 of credit, roughly ten keyword searches 45. Self-hosting removes the markup. You run one container image, supply a DataForSEO key built by base64-encoding your email and password, and the app starts with authentication switched off — the guide says to keep it behind your own proxy or on your own machine 3.
I ran that path. Cloning the repository and installing dependencies took about twenty seconds and its 54 database migrations applied cleanly, but the documented development server refused to start: the prerequisites say Node 20 and up, and Node 22.13 was not new enough for the Cloudflare plugin it runs on. On Node 24 it came up and served the app. Self-hosting is genuine, and it is an afternoon if your toolchain cooperates 3.
Two things to weigh before switching. The data has a ceiling: it is DataForSEO's index, not Ahrefs'. An independent review priced the same jobs inside its own DataForSEO account and found keyword and backlink depth is where you feel the difference, more than the markup — its full test cost about 70 cents of direct requests, and the same requests through the hosted layer would be about 89 6. That reviewer sells a competing SEO product and earns affiliate fees from DataForSEO, and had not used OpenSEO when he wrote it, so weigh it accordingly. Second, the project is pre-1.0: the repository has been public since February 2026, the newest release was version 0.1.9 in mid-September, and almost all of the contribution is one person's 6.
Opengeni: the code is free, the platform isn't
Opengeni is infrastructure for putting an agent inside your own product: sessions that survive a worker restart and resume where they stopped, a sandbox so agent code never runs beside your secrets, per-user OAuth credentials, tools and MCP wiring, memory, multi-tenancy with row-level security, and React components so the agent appears inside your interface under your brand 13. The open-source grant is complete — API, console, workers, Helm chart and reference Terraform are all Apache 2.0 7. Cloud is priced as model cost plus 5%, with no seat fee, platform fee or minimum commitment 8.
Self-hosting is where the second bill appears. The same page lists what you operate: the API, the web console, two Temporal workers, Postgres with pgvector, Temporal, NATS, object storage, and a sandbox backend. Installation is a Helm chart on Kubernetes with reference Terraform for AWS, Azure and GCP. In production the docs expect managed services for Postgres, Temporal, NATS and object storage, because the chart's bundled copies are for local testing. Before you expose it you need a gateway with TLS, authentication on every user-facing route, rate and request-size limits, and streaming timeouts of at least an hour. Some upgrades require stopping every old API and worker first, and the old image must not be restarted afterwards 7.
That is the trade: 5% to have the platform run for you, or a platform team's standing appointment. This is the one of the three I could not test — standing up Kubernetes is beyond what my machine can do, so this description comes from their documentation rather than from my own use of it 7.
AUDR: the standard is free, the measurement isn't
AUDR is a different kind of free: not software that does a job, but the shape of a record, published as an open standard with an Apache 2.0 licence 910. It defines one JSON record per metered operation. A run id is minted by whatever starts the run, passed through the router and echoed back, so every system that touched it carries the same identifier. Each fact has one owner: the harness knows the customer and the environment, the router knows the tokens. A sink joins records that share a run and span id, conflicting records are rejected rather than merged, and a correction is issued as a new record rather than an edit 14.
What it is aimed at is real: a run can be fully observable at every layer and still leave you with no single record of what that customer cost you last month 9. This is the one I ran. I installed the Python SDK and emitted three records from a single run — a model call from the harness, the same span as seen by a router, and a tool call. All three validated against the published version 1.0.0 schema, and the SDK rejected a badly formatted provider name at the moment the record was built 10.
The part you still pay for is the part the standard deliberately leaves out. A record carries counts and attribution, not money: pricing and rating sit outside AUDR, so you still need something downstream that turns tokens into a number. What landed in my file was one line per record, because joining those into one record per run per customer is the sink's job, not the SDK's. And a record I built with no attribution at all was accepted — nothing forces a record to say whose cost it is, so a harness that never sets the customer collects records nobody can bill for 14.
Five adapters ship today — NVIDIA NeMo Relay, LiteLLM, Merge Gateway, Vercel AI SDK and Mastra — with two sinks, Chargebee and Lago, and OpenRouter in development; a runtime outside that list means writing the adapter, which they estimate at about 200 lines against shared fixtures. The standard is designed to sit on top of OpenTelemetry rather than replace it, adding the required fields and delivery rules that observability leaves open, and it is explicit that it complements the FOCUS billing schema by covering the upstream half 910.
The startup-job picks
For cheaper SEO data inside the agent you already use, start with OpenSEO. Ten dollars, one afternoon, and watch the usage line rather than the subscription line 4.
For an agent inside your product, with nobody on the team wanting to run a workflow engine, start with Opengeni on the cloud and pay the 5%. Move it in-house once the cluster already exists, or when the rule is that this data never leaves your cloud 8.
For knowing what your agent costs per customer before you bill for it, start with AUDR. Write records to a plain file first, and see whether the numbers land where you assumed they would 9.
Open source changed what you pay for, not whether you pay. All three hand you the code and move a bill onto your team instead: a data vendor's invoice, a platform team's time, or the evening you spend instrumenting every layer a run passes through. Don't pick by which one is finally free. Pick the bill your team can actually hold, and hand the rest to someone who charges you for carrying it.
Two limits on what you have just read. First, this is a public-source buying screen rather than a logged-in trial inside each product: the setup paths, prices and limits come from the vendors' own pages and documentation, plus what the discovery boards recorded about when each product launched. Where I did run something, I say so in the section. Second, verify runtime behaviour on one reversible workflow of your own before any of these sits in front of paying customers.
References
- 1Product Hunt weekly leaderboard, week of October 5 2026
producthunt.com
- 2Product Hunt daily leaderboard, October 8 2026
producthunt.com
- 3OpenSEO Docker self-hosting
github.com
- 4OpenSEO pricing
openseo.so
- 5OpenSEO repository
github.com
- 6OpenSEO review
ryandoser.com
- 7Opengeni self-hosting
docs.opengeni.ai
- 8Opengeni pricing
opengeni.ai
- 9AUDR
openaudr.dev
- 10AUDR repository
github.com
- 11OpenSEO
openseo.so
- 12OpenSEO agent setup
openseo.so
- 13Opengeni
opengeni.ai
- 14AUDR specification v1.0.0
openaudr.dev
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.