# Pay for value, not for data Every plan includes unlimited data volume and unlimited queries. You pay for the people who use DataFocal and the collectors that feed it, never for how much they send. ## Open source: Free Run the engine yourself. - The full engine, Apache 2.0 license - Findings, drill-down and search - MCP server for your agents - Your own storage - Community support ## Team: $30 per user / month Managed by us, ready in minutes. - Everything in Open source - Hosted and upgraded for you - Unlimited queries, including from agents - Collectors from $12 / month, less at volume - Managed storage, or bring your own bucket ## Enterprise: Custom For large teams and scale. - Everything in Team - SSO and roles - Multiple teams and business units - Long-horizon AI analysis - Dedicated or in-your-cloud deployment - Support SLA and onboarding ## Collectors A collector runs where your data is and indexes it there. It costs the same whether it sends a gigabyte a day or a hundred. - First 250: $12 per collector / month - 251 to 1,000: $9 per collector / month - Above 1,000: $6 per collector / month - Collectors on developer machines, reading AI agent transcripts, are free. - Billed on the month's average count, not the peak, so autoscaling doesn't cost you. ## Storage: managed, or your own bucket Either way there's no charge for data volume. ### Managed by us: Cost + 20% We run storage for you on low-cost object storage and pass the bill through at cost plus 20%, which covers running it. - About $0.016 per GB-month at today's prices - Example: 100 GB a day kept 30 days is about $54 / month - Nothing to set up ### Your own bucket: $0 Point DataFocal at an S3-compatible bucket in your own account. The data never leaves it, and you pay your cloud provider directly. - AWS S3, Cloudflare R2, MinIO or any S3-compatible store - Your retention, encryption and access policies - Available on Team and Enterprise ## Compare plans | Feature | Open source | Team | Enterprise | |---|---|---|---| | Data volume | Unlimited | Unlimited | Unlimited | | Queries, human and agent | Unlimited | Unlimited | Unlimited | | Collectors | Unlimited | $12, $9 or $6 each by volume; developer machines free | Volume pricing | | Storage | Your own | Managed at cost + 20%, or your bucket at $0 | Managed, your bucket or your cloud | | Hosting | You run it | Managed | Managed, dedicated or your cloud | | SSO and roles | — | — | ✓ | | Support | Community | Email | SLA and onboarding | ## Questions ### Why don't you charge for data volume? Because it punishes you for looking. DataFocal indexes data where it's collected and stores it compactly, so more data costs us little. You pay for the people and collectors that use DataFocal; storage is passed through at cost plus 20%, or free in your own bucket. ### What counts as a collector? Each running collector that sends data to DataFocal, usually one per server, virtual machine or Kubernetes node. Collectors on developer machines that read AI agent transcripts are free. ### Can my agents use DataFocal? Yes. DataFocal is an MCP server: connect Claude Code, Codex, Cursor or any MCP client. Agent queries are unlimited, like everyone else's. --- # What is DataFocal DataFocal is observability that starts with AI. It reads what your AI agents, services and devices already record, learns what normal looks like, and shows you what changed, why, and the runs and events behind it. ## Find, explain, focus - **Find.** DataFocal brings what stands out to the top: spend, cache hits, context size, failures and outcomes, across every agent, model and tool. - **Explain.** Every number comes with its why: what fills the context window, how it grows across a run, and where the cost goes. - **Focus.** Narrow to the one run, call or event behind a number, with the evidence attached. ## What it reads - Claude Code and Codex transcripts - LLM gateway logs and OpenTelemetry GenAI traces - Application logs and traces, and Kubernetes - Robot and edge device events ## How it's priced Per seat, with no charge for data volume or queries. See [pricing](/pricing). --- # Quickstart See where your AI agents' spend goes in about five minutes. You'll install the collector where your agents run, and DataFocal reads the transcripts they already write. ## 1. Install Run it on a machine where Claude Code or Codex runs. It finds their transcripts on its own; nothing to instrument. Pick one: ### Free, on your machine Everything runs locally: the collector, storage and console. No account needed. You need Linux with systemd and docker, plus Go and Node to build it. ```sh git clone https://github.com/datafocal/datafocal cd datafocal make install-local ``` The console is then at http://localhost:8080/ui/. ### Team, hosted by us Sends to your DataFocal workspace so your whole team sees it. [Talk to us](/contact) and we'll set up your workspace and collectors with you. ## 2. Open the console The AI view fills in with spend, cost per run, cache hit rate and outcomes from your existing transcripts. Findings appear as soon as DataFocal has enough history to know what normal is. [Image: The DataFocal AI view overview] ## 3. Ask what changed Type a question in the console, or ask your agent once it's [connected](/docs/connect-an-agent). Searching, filtering and drilling down are always free. --- # Connect an agent DataFocal is an MCP server. Once it's connected, your agent can ask what changed and why, and gets findings with the evidence attached. The commands below use a local install's address. A Team workspace gives you its own address to use instead. ## Claude Code ```sh claude mcp add --transport http datafocal http://localhost:8080/mcp ``` ## Codex ```sh codex mcp add datafocal --url http://localhost:8080/mcp ``` ## Any MCP client ```json { "datafocal": { "url": "http://localhost:8080/mcp" } } ``` ## What to ask - "Where does our agent spend go?" - "Why did that run fail?" - "What changed since the deploy?" - "Show me the raw events behind that number." Queries from agents are unlimited, like everyone else's.