Where Tessera fits

The single differentiator: Tessera is the only entry on this page that returns a smaller bill, not a better dashboard. Every other category gives you better visibility (which we also do) but leaves the actual savings work to you.

Honest categorical fit. Tessera is wrong for you if any of these apply:

• You need on-prem / air-gapped deployment today.Pick LiteLLM self-host or Langfuse self-host. Our enterprise "deploy to your Cloudflare account" tier is post-revenue (see hosted-only tradeoff).
• You need rich observability + traces as the primary product. Pick Langfuse (most mature OSS platform) or OpenLIT (OTel-native, GPU monitoring). Tessera ships per-request audit rows + CSV export but is not an observability platform — we have a dashboard, not a traces UI.
• You need 100+ provider coverage for unification.Pick LiteLLM or OpenRouter. We support 14 providers — enough for ~95% of production workloads but explicitly less broad than LiteLLM's 100+.
• Your stack is > 80% AWS Bedrock. Bedrock IAM/STS integration is on our roadmap, not shipped. Pick LiteLLM or LangChain Bedrock direct, route through us for the non-Bedrock portion. Talk to us if Bedrock priority is high — we will accelerate if the contract justifies.
• You need OSS evaluation infrastructure.Pick promptfoo standalone — it ships the evaluation framework we use internally for our quality canary. promptfoo is part of OpenAI now and stays the canonical choice for "test your prompts vs golden set."

Tessera is the right tool when all of these apply:

1. You spend $3-100K/month on OpenAI / Anthropic / Mistral / Cohere / Groq combined. Below $3K the absolute savings rarely justify the integration time; above $100K you are usually a candidate for our direct Enterprise sales process.
2. You want the bill cut at request time, not after a budget review. The observability platforms show you waste; Tessera eliminates it.
3. You can tolerate 15-25 ms p50 latency overhead from the Cloudflare edge hop (negative on cache hits — 10-50× faster).
4. You want the vendor incentive aligned with cost outcome. Flat monthly price by token volume — you keep 100% of savings — different incentive shape from per-token gateway fees.
5. You want every billable line auditable. See /trust for the two-engineers-three-hours re-derivation guarantee.

Tessera is designed to co-exist with the observability and gateway tools, not replace them. Recommended topologies:

• Tessera + Langfuse: Tessera routes + optimizes; Langfuse captures the resulting traces. Our HTTP response includes the standard OTel gen_ai.* spans plus our mechanics_stack header so Langfuse sees the optimization context.
• Tessera + Helicone:Same shape as Langfuse. Helicone captures requests; we mutate them. Helicone's prompt-management v2 for A/B prompts works upstream of our routing; we honor whatever prompt you send.
• Tessera + Promptfoo: This is our internal stack. We use Promptfoo for the quality canary that gates every mechanic. You can use Promptfoo separately for your release-gate test suite — same eval set feeds both.
• Tessera behind LiteLLM:Workable if you need LiteLLM's broader provider unification. Point LiteLLM's upstream at Tessera for the providers we support; bypass us for the rest. We only price the gross tokens routed through us, so this stays incentive-clean.
• Tessera behind Portkey: Workable but unusual. Portkey charges a per-token gateway fee on top of provider cost; we save against the underlying provider rate. Stack is fine but you pay both.

The OSS alternatives (Langfuse, OpenLIT, LiteLLM self-host) are excellent and we recommend them in their category. The honest reasons to add Tessera on top:

• The bill comes back smaller.Self-hosted observability tells you where the money goes; it does not reduce the bill. Our 10 mechanics return measured savings in the 30-60% range on representative workloads — and you keep 100% of it. The flat monthly subscription is a small fraction of a typical month's savings.
• The mechanic implementations are non-trivial to clone. Auto-route with eval gate, per-stack canary auto-rollback, multi- source pricing catalog with 3-source verify, M11 cross-provider failover with bounded data residency — these are 6-12 months of engineering work per mechanic if you build in-house. We absorb that cost across all customers.
• Quality SLA is contractually committed. If a mechanic drops your stack mean-score below 0.95 for 3 consecutive days, the auto-rollback fires AND you get a 10% SLA credit. No OSS tool exists that ships this enforcement loop as a service.

The OSS direction we WOULD recommend if you fundamentally want to build it yourself: take LiteLLM as the gateway base, Langfuse as the observability backend, Promptfoo as the eval framework, plug in your own routing + canary + rollback logic. That stack works; it just costs more in engineering time than the flat monthly subscription covers for most workloads.

If you are evaluating Tessera against a specific alternative and want our honest take on which is the better fit for your stack — email contact@tesseraai.io with the workload shape (provider mix, monthly spend, primary use case). We answer in our own voice, not a sales pitch — sometimes the answer is "the alternative fits you better, here is why."