Know what your agents did.
Prove it when it matters.
Self-hostable AI agent observability: trace ingestion, evals, per-model cost attribution, and tamper-evident audit logs. One binary. Your infrastructure. No per-seat tax.
One agent run. Every span accounted for.
Each step of the run as Spanwise records it — the model behind it, the tokens it used, what it cost, and how long it took.
The per-seat observability bill is broken
LangSmith Plus costs $39–$49 per seat per month. A 20-person team pays $780/mo before any overages — and that is before usage overages, which is where the real bill lives at ~1M traces/mo.
Meanwhile the independent vendors keep getting acquired: ClickHouse bought Langfuse (Jan 2026), Dynatrace bought Arize for $915M (Oct 2026), and Helicone is in maintenance mode. Your observability roadmap shouldn't depend on someone else's acquisition strategy.
Spanwise Team: $49/mo flat. Unlimited seats. 10M traces/mo included.
How it works
- Point your pipeline at us
Send OTLP/HTTP-JSON or the flat span format to POST /v1/traces with your workspace key. - Watch traces and costs
List and inspect traces in the viewer API; per-model token spend is attributed automatically. - Run evals
Score agent outputs with the built-in eval harness — heuristic judge for CI, pluggable LLM judge. - Export the evidence
Pull the hash-chained audit log any time and verify it independently.
Why Spanwise
Flat per-workspace pricing
Unlimited seats. A 20-person team costs $49/mo flat — a 16× headline difference vs. per-seat plans, stated without overages.
Single-binary self-host
One binary, your infrastructure, SQLite to start / Postgres when you scale. No vendor cloud required. Shipped: the 0.1.0 one-binary release, SHA-256 checksummed (see Docs).
Tamper-evident audit logs
Hash-chained span/audit exports mapped to EU AI Act Article 12 logging obligations — sold as compliance readiness, not certification.
OTel-compatible ingestion
Accepts OTLP/HTTP-JSON subset plus a flat span shape; GenAI semantic-convention attributes honored.
Questions, answered honestly
- Is this a LangSmith replacement?
- For teams whose core needs are tracing, evals, and cost attribution — yes, at a fraction of the price. We don't claim feature parity on every LangSmith surface; we claim the specific wedges: price, self-host, audit logs.
- Do you certify EU AI Act compliance?
- No. We give you tamper-evident logs mapped to Article 12 record-keeping expectations — readiness, honestly labeled. Fines are not currently in effect and we never claim otherwise.
- Where does my data live?
- On your infrastructure if you self-host (Enterprise). SQLite by default, Postgres when you scale.
- What does the eval harness do?
- Scores agent outputs against expected key points. The heuristic judge is deterministic and runs in CI; the LLM judge is a real integration point that requires your provider key — it refuses to run without one rather than faking a score.