How much does Respan cost?

Free covers the full platform with 100,000 logs, 1,000 scores, 5 datasets, 2 evaluators, 10 prompts and 7 day retention. Team is $199 a month billed yearly with 10,000 scores, unlimited datasets, evaluators and prompts, 30 day retention, 5 members and $15 per extra member. Extra logs are $8 per 100,000 and extra scores $1 per 1,000. Enterprise is quoted with SAML SSO, self hosting, custom SLAs and a HIPAA BAA. Gateway tokens bill at Respan's bulk rates with no fee, or on the customer's own provider keys.

What are the best alternatives to Respan?

Based on documented capability coverage across the Agentic Index 14 point taxonomy, the closest alternatives to Respan among agent infrastructure platforms are OpenRouter, Composio, Inworld AI, Maxim AI. Each is graded on the same public evidence, and no vendor pays for placement.

Does Respan support MCP and tool calling?

Full under the Agentic Index verification standard. Respan publishes public evidence of Model Context Protocol (MCP) support, tool calling, and API or SDK extensibility for connecting agents to external tools. Evidence basis: A REST API reference covers the gateway itself plus spans, traces, prompts and versions, routes, limit policies, behaviors, custom models and providers, caches, metrics, users and temporary API keys, and traces can also be ingested over OTL. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Does Respan have SOC 2, SSO and role based access control?

Full under the Agentic Index verification standard. Respan publishes public evidence of enterprise security and identity controls, covering a compliance posture such as SOC 2 or ISO 27001, single sign on (SSO or SAML), and role based access control (RBAC). Evidence basis: Projects carry four roles: Owner, Admin, Member and Annotator, with only owners and admins able to invite people, change roles or delete the project, and annotators limited to review work. API keys are scoped to a project and environment wi. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Can Respan be self hosted or run in a private VPC?

Full under the Agentic Index verification standard. Respan publishes public evidence of self hosted or private VPC deployment, along with data residency options for controlling where agent data is processed and stored. Evidence basis: The gateway can run entirely on the customer's own infrastructure: the open source Respan Gateway ships under the MIT or Apache 2.0 license as one binary, a Python package that runs as a server or inside the application, or a Docker image,. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Back to vendors

Respan

Also known as: Keywords AI, keywordsai.co, respan.ai, Respan Gateway

Visit site

Agent infrastructureindependentVerified 2026-11-09

Entry priceFree to start with 100,000 logs and 1,000 scores. Team is $199 a month billed yearly; Enterprise is quoted. Gateway tokens bill at Respan's bulk rates with no fee, or on the customer's own keys.Full pricing detail

Respan, formerly Keywords AI, is an AI gateway that puts more than 1,000 models behind one API key with fallbacks, load balancing, caching, PII redaction and budgets per key or customer. Every call becomes a traced span, live traffic is scored against built in and custom behaviors, and red team campaigns attack an agent and grade the result. The gateway core is open source. A free plan covers 100,000 logs, Team costs $199 a month billed yearly, and Enterprise adds SAML and self hosting.

Respan is the brand of Keywords AI Inc. of Alameda, California, which renamed itself from Keywords AI in 2026 and went through Y Combinator's Winter 2024 batch. It sells one platform for teams shipping LLM applications and agents: an AI gateway, observability, evaluation and security testing, with an open source gateway under the MIT or Apache 2.0 license.

The gateway takes OpenAI, Anthropic and Gemini SDK calls at one endpoint and routes them to more than 1,000 models from providers such as OpenAI, Anthropic, Google, AWS Bedrock, Vertex AI, Azure, Mistral, DeepSeek and Groq, using Respan's bulk rates with no fee or the customer's own provider keys. Routes and aliases, weighted load balancing, provider routing and fallbacks keep requests flowing, caching cuts repeats, and PII redaction masks personal data before it reaches a provider. In the open source gateway, a tool loop lets any chat model call tool servers, including MCP servers and wrapped APIs such as Tavily, Exa, Browserbase, Daytona and Zep, inside one request, with allow and deny rules and a step cap.

Every call is logged as a span with cost and latency, nested into traces and threads by agent step, tool run and retrieval, and teams can query logs in SQL, plain English or from a CLI or the Respan MCP server. Behaviors score each new span for refusals, instruction misses, incomplete answers and completed tasks, plus definitions the team writes, and red team campaigns send prompt injection, system prompt extraction and excessive agency attacks to an agent and return a graded report that can be rerun after a fix. Monitors alert on errors, cost and user frustration.

The Free plan includes the full platform with 100,000 logs, 1,000 scores, 2 evaluators and 7 day retention. Team costs $199 a month billed yearly with 10,000 scores, unlimited evaluators, 30 day retention and 5 members plus $15 for each extra seat. Extra logs cost $8 per 100,000 and extra scores $1 per 1,000. Enterprise is quoted and adds SAML single sign on, self hosted deployment, custom SLAs and a HIPAA BAA. Projects carry Owner, Admin, Member and Annotator roles, API keys are scoped by environment, IP range, expiry and spend, and Respan holds a SOC 2 audit and ISO 27001 certification, with zero data retention agreements on request.

Respan fits engineering teams that want routing, cost control, tracing and quality scoring for their agents in one service, and teams that want to run the gateway themselves. It fits less well for buyers looking for an agent builder, prebuilt agents or a knowledge base, or for those that need hosted data in a chosen region without moving to self hosting.

Vendor details

Canonical URL

https://www.respan.ai

Subcategory

AI gateway with built in tracing, live behavior scoring and red teaming for agents, hosted or self hosted from an open source core

Funding status

Privately held. Keywords AI Inc., doing business as Respan, went through Y Combinator's Winter 2024 batch.

Company status

independent

Use cases & customers

Primary use cases

routing calls to more than 1,000 models through one endpoint with fallbacks, load balancing and cachingsetting budgets and rate limits per API key and per end customertracing agent runs span by span with cost, latency and errorsscoring live traffic against built in and custom behaviorsrunning red team campaigns against an agent and retesting after fixes

Target customers

engineering teams running LLM applications and agents in production that need routing, fallbacks and cost control across many model providersAI product teams that want every agent step traced and live traffic scored for quality issuessecurity and platform teams that want to red team their agents before releasecompanies that need to self host the gateway or keep request content out of vendor logs

Deployment options

Respan Cloud (Free and Team)Self hosted platform on EnterpriseOpen source Respan Gateway as a binary, Python package or Docker image

Integrations

Gateway endpoints for the OpenAI, Anthropic and Gemini SDKs and more than 30 model providers including OpenAI, Anthropic, Google Gemini, xAI, AWS Bedrock, Vertex AI, Azure OpenAI, Mistral, DeepSeek, Groq, Cohere and Together AI; tracing for OpenAI Agents, Claude Agent SDK, Vercel AI SDK, LangChain, LangGraph, CrewAI, LlamaIndex, Pydantic AI, Google ADK and Microsoft Agent Framework plus native OpenTelemetry; coding agents such as Claude Code, Codex CLI and Cursor; open source gateway tool servers for Tavily, Exa, Linkup, Apify, Browserbase, Daytona, Reducto, Zep, Cloudinary and ElevenLabs; Respan MCP server and CLI; PostHog and webhook exports.

In practice

A support startup sends every model call through Respan with a budget on each of its customer accounts. When one customer's usage hits its cap, Respan returns a 402 for that customer only, while fallbacks keep everyone else running through an Anthropic outage by switching to Bedrock.

An AI team writes a custom behavior that flags replies promising refunds outside policy. Respan scores every new span against it, the team opens the spans behind a spike, fixes the prompt and watches the behavior's rate drop the next day.

Before launch, a fintech connects its agent to Respan's red team CLI. The first campaign gets 6 of 48 attacks through, mostly prompt injection, and after the team adds a guardrail the retest blocks all 48.

Sources & related URLs

Related / legacy domains

https://www.respan.ai/docs/documentation/overview https://www.respan.ai/pricing https://platform.respan.ai https://github.com/respanai/respan-gateway https://trustcenter.respan.ai/ https://status.respan.ai

Research sources

Agentic Index coverage score

10.0 / 14 capabilities · 71%

Integrations & Tool Calling	Full
The open source gateway runs tool calls for any chat model. Each tool server is a remote MCP server, a local one started as a command, or an outside API the gateway wraps as MCP tools with the team's own key: Tavily, Exa and Linkup for search, Apify and Browserbase for the web, Daytona for code sandboxes, Reducto for documents, Cloudinary for media, Zep and ElevenLabs among them. With the tool loop on, the gateway executes the calls a model makes, such as creating a sandbox or deleting an asset, and returns one reply. Sourcerespan.ai/docs/oss-gateway/toolsread 2026-11-09
Workflow Orchestration	Partial
A single request can run several steps. With the tool loop on, the gateway offers tools to the model, runs the calls it makes, up to four at once, and calls the model again with the results until it answers or reaches the step cap, eight model calls by default, within a five minute limit. Routes chain a primary model to fallbacks and weighted alternatives. The agent's own plan, its branching and any hand off between agents stay in the team's code or framework. Sourcerespan.ai/docs/oss-gateway/toolsread 2026-11-09
Knowledge Grounding & RAG	Not documented
Respan routes, caches and records model calls rather than holding a company's documents. Embedding and file requests pass through to providers, cached responses replay identical requests, and in the open source gateway Reducto parsing and web search tools return content at request time. No index, vector store or knowledge base over the customer's content is maintained for agents to query, so retrieval stays in the customer's own stack, where Respan traces each retrieval step. Sourcerespan.ai/docs/ai-gateway/features/cachingread 2026-11-09
Human Oversight & Guardrails	Partial
Limits set by the team stop requests rather than a person signing off on them. Each API key can carry a monthly spending cap, a lifetime request cap and request and token rate limits, limit policies cap cost or tokens per window, and each end customer can get its own budget, with calls over a cap refused. PII redaction masks 17 kinds of personal data before a provider sees it, and the tool loop's allow and deny lists keep tools such as deletes out of a model's reach. Annotators review outputs after the fact. Sourcerespan.ai/docs/ai-gateway/spend/budgets-and-limitsread 2026-11-09
Security, Identity & Governance	Full
Projects carry four roles: Owner, Admin, Member and Annotator, with only owners and admins able to invite people, change roles or delete the project, and annotators limited to review work. API keys are scoped to a project and environment with their own permissions, IP ranges, expiry and spend caps. Google sign in comes on every plan and SAML single sign on on Enterprise. Respan holds a SOC 2 audit with the report on request and ISO 27001 certification, signs HIPAA BAAs, runs a trust center and signs a DPA. Sourcerespan.ai/docs/documentation/admin/collaborateread 2026-11-09
Observability & Auditability	Full
Each gateway call becomes a span with its input, output, tokens, cost and latency, nested into traces by agent turn, tool run and retrieval and grouped into threads per end customer, and agent frameworks or OpenTelemetry can send spans from code that bypasses the gateway. Teams search logs, open debug traces, ask questions in SQL or plain English, and run the same queries from a CLI or the Respan MCP server. Error groups, incidents, metrics dashboards and scheduled reports round it out. Sourcerespan.ai/docs/ai-gateway/observability/logs-and-tracesread 2026-11-09
Memory & State Persistence	Partial
Memory comes from a partner rather than from Respan. The open source gateway wraps Zep as a tool server, so a model in the tool loop can add messages and data to Zep and read back graph and thread context across sessions on the team's own Zep key. The hosted platform groups a conversation's calls into threads by customer and thread ID for review, and the Responses API's stored items pass through to providers, but Respan keeps no memory of its own for an agent to read. Sourcerespan.ai/docs/oss-gateway/toolsread 2026-11-09
Deployment & Data Residency	Full
The gateway can run entirely on the customer's own infrastructure: the open source Respan Gateway ships under the MIT or Apache 2.0 license as one binary, a Python package that runs as a server or inside the application, or a Docker image, and Enterprise adds a self hosted deployment of the platform. On Respan Cloud, logging can be turned off per request or for the organization, retention runs 7 or 30 days or a custom window, and zero data retention agreements keep providers from storing content. Sourcerespan.ai/docs/oss-gateway/introductionread 2026-11-09
Prebuilt Agents / Templates / Packs	Not documented
Respan ships no ready made agents, agent templates or installable packs. The prompt library holds the team's own versioned prompts with Jinja templates, and the open source gateway's examples are single runnable files for plugins such as guardrails, PII redaction across streams, billing and admission control. The hosted sample agent in red teaming is a deliberately weak target for trying a campaign, and the built in behaviors are scoring presets rather than agents a team adopts. Sourcerespan.ai/docs/oss-gateway/pluginsread 2026-11-09
Triggers & Channel Coverage	Full
Scoring starts on its own as traffic arrives: once behaviors are on, every new span is scored against each active behavior with no request from the team. Monitors watch errors, cost, latency and user frustration and send alerts when a threshold is crossed, error groups open incidents, and webhooks and scheduled exports push logs out as often as every 15 minutes. Gateway requests themselves arrive over the OpenAI, Anthropic, Gemini and realtime endpoints whenever the application calls. Sourcerespan.ai/docs/documentation/features/monitoring/monitorsread 2026-11-09
Model Flexibility & Routing	Full
The team picks any of more than 1,000 models per request from providers including OpenAI, Anthropic, Google, xAI, Bedrock, Vertex AI, Azure, Mistral, DeepSeek, Groq and Together AI, and the Anthropic SDK can call GPT models while the OpenAI SDK calls Claude. Routes and aliases name a model once, load balancing splits traffic by weight across models, providers and keys, provider routing pins or orders providers, and fallbacks switch on errors. Custom providers and models add anything else that speaks a supported format. Sourcerespan.ai/docs/ai-gateway/routing/provider-routingread 2026-11-09
APIs / SDKs / MCP Extensibility	Full
A REST API reference covers the gateway itself plus spans, traces, prompts and versions, routes, limit policies, behaviors, custom models and providers, caches, metrics, users and temporary API keys, and traces can also be ingested over OTLP. The respan CLI queries logs and traces from a terminal, and the Respan MCP server at a published endpoint signs in with OAuth or a bearer key so Claude, ChatGPT, Cursor or Codex can query and manage the account. Sourcerespan.ai/docs/documentation/mcpread 2026-11-09
Testing, Debugging & Optimization	Full
Behaviors score every new span from live traffic: built in ones flag refusals, instruction misses, incomplete answers, structured output and completed tasks, and teams write their own in plain language, each firing above 65 percent confidence with a verdict a person can override. Red team campaigns send prompt injection, system prompt extraction and excessive agency attacks to the team's own agent through a CLI, grade how many get through by category and can be rerun after a fix. Datasets, LLM judges and annotation queues sit alongside. Sourcerespan.ai/docs/documentation/features/pulse/behaviorsread 2026-11-09
Browser / Computer-use	Partial
Browser work runs on other companies' engines that the open source gateway wires in as tools. With a Browserbase key, a model in the tool loop can search, fetch pages, start a browser agent run and get a live view; TinyFish adds a browser agent of its own, and Apify can browse the web and run scraping actors. Respan hosts no browser or desktop itself, and on the hosted gateway, computer use tools from providers pass through for the customer's own code to carry out. Sourcerespan.ai/docs/oss-gateway/toolsread 2026-11-09

The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded

Pricing

Free to start with 100,000 logs and 1,000 scores. Team is $199 a month billed yearly; Enterprise is quoted. Gateway tokens bill at Respan's bulk rates with no fee, or on the customer's own keys.

Platform plan billed yearly, plus logs and evaluation scores past the plan allowance, extra seats on Team, and model tokens through the gateway

Free tierTrial available

Included quota

Free includes 100,000 logs, 1,000 scores, 5 datasets, 2 evaluators, 10 prompts, 1 workspace, 4,200 proxy requests a minute and 7 day retention. Team includes 10,000 scores, unlimited datasets, evaluators and prompts, unlimited workspaces, 5 members, 8,400 proxy requests a minute and 30 day retention.

What is public

Plan prices and allowances for Free and Team, per unit prices for extra logs, scores and seats, throughput limits, retention windows and the security features each plan adds.

Billing mechanics

Team bills yearly with included logs and scores, and usage past the allowance bills per 100,000 logs and per 1,000 scores; gateway tokens draw from prepaid credits at Respan's rates or go to the customer's own keys.

Cost watchouts

High traffic agents can pass the log and score allowances quickly, HIPAA compliance is listed as a $500 a month add on, and model tokens bill separately from the platform plan.

Variable cost rationale

The platform plan is fixed, but logs and evaluation scores past the allowance bill per unit, and gateway tokens bill by usage at Respan's rates unless the team brings its own provider keys.

Additional watchouts

Behaviors need the Team plan or higher, and SAML single sign on and self hosting of the platform need Enterprise.

Overage / add-ons

Logs past the allowance cost $8 per 100,000 and scores $1 per 1,000, with volume discounts on Enterprise; extra Team members cost $15 each.

Sales call required

Mixed (some tiers require a call)

Free / trial

Free plan with the full platform, 100,000 logs, 1,000 scores, 5 datasets, 2 evaluators, 10 prompts, unlimited seats and 7 day retention, no card required; Team can be tried for free.

Lowest paid plan

Team at $199 a month billed yearly

Commercial notes

Respan, formerly Keywords AI, sells the gateway, observability, evaluation and red teaming as one platform, with the gateway core also open source for self hosting.

Key ambiguities

The Team price is shown with a 17 percent yearly discount and no monthly list price, and HIPAA appears both as an Enterprise feature and as a $500 a month add on.

Cancellation / refund

Free needs no card; Team and Enterprise are billed yearly under the terms of use.

Support SLA / resale

Free has a community Discord, Team adds email support, a private Slack channel and an 8 hour support SLA with a 99.99 percent uptime SLA, and Enterprise adds priority support, custom SLAs and a dedicated support engineer.

Missing data

Enterprise prices, the monthly list price for Team and hosted data regions.

Agentic Index verified 2026-11-09

Official pricing page Docs

Related vendors

Alternatives to Respan

The closest documented capability profiles to Respan among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.

OpenRouter10.5 / 14Fuller documented coverage on Human Oversight & Guardrails
Composio9.5 / 14Fuller documented coverage on Human Oversight & Guardrails
Inworld AI8.5 / 14A lighter documented profile than Respan
Maxim AI8.5 / 14A lighter documented profile than Respan
TrueFoundry9.5 / 14Fuller documented coverage on Human Oversight & Guardrails
Albato10.0 / 14Adds documented Prebuilt Agents, Templates & Packs

Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded

Compare Respan with other vendors →Compare all agent infrastructure platforms →See where Respan ranks →

Data confidence: medium