Credal

• Microsoft, Glean, and other platform vendors could fold agent registries and permission-aware MCP access into their own products, which would shrink the set of buyers who need a separate Credal gateway.
• If customers standardize their agents on one AI vendor’s tools, Credal’s case for one gateway that applies the same rules across Claude, ChatGPT, Cursor, and Slack weakens.
• Salesforce now ships hosted MCP servers and an Agent Fabric registry, according to Credal’s own Checkr case study, so vendors that hold a customer’s system of record could govern agent access to it without a separate gateway.

Credal sells to enterprises whose AI agents need governed access to internal systems such as Salesforce, Snowflake, Jira, and Confluence. Its MCP Platform lets each department publish its own governed MCP server, the interface AI tools such as Claude, ChatGPT, and Cursor use to call those systems, without rebuilding authentication, permissions, and auditing each time.

The buyer Credal names is the platform team. Ravin Thambapillai told the Frontlines podcast that the primary customer profile is the platform engineering lead at technology companies of 1,000 to 10,000 people, and Credal’s registry page describes every agent in an organization versioned, governed, and auditable in one place.

Credal also frames the problem as cost and accuracy. Its homepage argues that over-provisioned agents are expensive and inaccurate because they carry too much context, and its Checkr case study reports AI answers on Salesforce data rising from 13% to 86% accurate. Both framings come from Credal’s own pages. No reviewed independent source quantifies the pain for this segment. [s10, s15, s2, s1, s14]

Credal sells three connected products. The MCP Platform is a gateway between agents and company systems that turns connected tools into department-specific MCP servers. The Agent Builder composes agents in a no-code canvas or through an API, and the Agent Registry versions and audits every agent, including agents built on third-party frameworks.

Credal applies its governance at the gateway. It inherits each user’s source-system permissions and enforces them at query time, scopes servers, tools, and actions to designated users, groups, or caller agents, and can require human approval on sensitive write actions. It also detects and redacts or blocks sensitive data before it leaves Credal or the customer’s VPC, and it exports audit logs to a SIEM through a native Splunk integration and OpenTelemetry.

Credal also sells cost control. Its consolidated audit log attributes model spend to the team, workflow, and model that incurred it, and the platform recommends a cheaper model where its accuracy matches. The platform is model-agnostic, so a customer can connect its own Azure OpenAI deployment or self-hosted models, and the capability detail comes from Credal’s own product pages and documentation. [s10, s11, s2, s4, s3, s1]

Credal competes with platform vendors that ship their own agent governance. Microsoft described Agent 365 in a November 2025 post as a control plane that helps customers deploy, organize, and govern agents, with a registry and access control among its five capabilities. Glean documents a managed, remote MCP server that exposes its permission-aware enterprise context to tools such as ChatGPT, Claude, and Copilot Studio without bypassing its permission model.

Security vendors approach the same buyer from the AI-security side. WitnessAI describes one platform for securing and governing employees and agents, with control over agent tool access. F5 announced its intent to acquire CalypsoAI in September 2025 and describes CalypsoAI’s platform as bringing real-time threat defense and data security to enterprises deploying AI.

Credal’s Checkr case study describes alternatives including Claude’s native plugins and skills, Salesforce’s hosted MCP servers and Agent Fabric registry, and an in-house build. Credal’s stated difference is one gateway that every surface calls, Claude, ChatGPT, Cursor, or a customer’s own app, so the same permissions and instructions apply whichever model is in use. TechCrunch described Credal’s emphasis on compliance and security in 2023 as the company’s own framing, at least the way Fischer told it. [s17, s18, s19, s20, s14, s10, s1, s7]

Credal names its customers publicly. Its homepage and case-studies page list Wise, MongoDB, Lattice, Checkr, Altana, Gecko, Incident.io, IFRS, Comcast, Flatiron Health, and Guidewheel as enterprises and startups that trust it. Its Y Combinator launch post names MongoDB, Wise, Checkr, Lattice, Comcast, the US Federal Government (HHS), and the IFRS as production customers.

Credal describes two customer deployments in its case studies. Checkr’s RevOps team built three Credal MCP servers, one per Salesforce instance, and Credal reports AI answer accuracy on that data rising from 13% to 86%. Its Guidewheel case study describes a dedicated RevOps MCP server alongside governed read-only access for the rest of the company.

Federal spending data on USAspending.gov confirms the government relationship independently. HHS’s Administration for Children and Families funded a contract to Credal for a generative AI platform, signed in August 2025. HigherGov’s summary of the record gives a potential value of $6,199,649 and says the award was sole source, on the government’s stated basis that only one company can provide the product.

The motion is enterprise sales that starts with a time-boxed pilot on one team or use case. Fischer told TechCrunch in October 2023 that Credal had 11 customers, several on six-figure contracts, and Ravin told Frontlines in an interview it published in February 2025 that Credal was growing 20% to 30% a month. Both figures are the founders’ own. Credal does not disclose current revenue or customer count. [s1, s12, s16, s14, s24, s23, s4, s7, s15]

Credal’s founders come from Palantir. TechCrunch reported that Jack Fischer and Ravin Thambapillai worked there before founding Credal. Credal’s team page says Ravin spent seven years at Palantir building AI systems for financial, life-sciences, and public-health institutions, after roles at Google, THG, and GoCardless.

Jack led Palantir teams applying AI in areas from air travel to healthcare to US defense, according to the team page, and Y Combinator’s company profile lists him as Palantir’s US commercial hiring manager from 2020 to 2022. The reviewed sources show no prior exit or sustained publication record for either founder, so the team evidence is relevant senior experience.

The wider team is small. Credal describes a team drawn from Palantir, Google, and NASA, Y Combinator’s profile lists 14 people, and the team page includes a forward-deployed engineer and Cameron, who built a Credal-powered security bot while at MongoDB. [s7, s13, s21]

Credal publishes a security posture aimed at regulated buyers. Its pricing page states SOC 2 Type II certification, HIPAA-compliant configurations, and GDPR-ready data processing agreements. Its security page describes participation in the EU-US Data Privacy Framework, its UK Extension, and the Swiss-U.S. Data Privacy Framework, plus zero-data-retention or equivalent agreements with OpenAI, Anthropic, and Cohere.

Deployment choice is the other half of the trust case. Credal offers multi-tenant cloud, a managed single-tenant instance, a Credal-managed instance inside the customer’s own cloud account, and an air-gapped on-premises install on the customer’s Kubernetes. Bring-your-own-key model access keeps AI traffic in the customer’s tenant.

Every plan includes SSO, SAML, and SCIM, consolidated audit logs with SIEM export, and a dedicated support channel, according to the pricing page. Credal points security teams to its security page and a trust portal for documentation. [s4, s3]

Add analyzed competitors to compare them side by side with Credal.

Credal targets enterprises running AI agents across many tools, and it sells to the platform and AI teams that build them. Ravin Thambapillai named the platform engineering lead at technology companies of 1,000 to 10,000 people as the primary buyer, and Credal’s pricing separates builder seats for people who create agents from user seats for everyone who uses them.

Credal’s named customers include regulated and public-sector organizations. Fischer described Wise to TechCrunch as a publicly traded, regulated European enterprise, and the Y Combinator launch post adds the US Federal Government (HHS) as a production customer, alongside MongoDB, Checkr, Lattice, Comcast, and the IFRS. Federal spending data shows HHS’s Administration for Children and Families funding a Credal contract for a generative AI platform, signed in August 2025.

The entry point is narrow. Most customers start with a time-boxed pilot on one team or use case with success criteria defined upfront, and Credal says most teams are running within days.

Credal’s core capability is a governed gateway between AI agents and company systems. The MCP Platform turns connected systems into department-specific MCP servers, inherits source-system permissions, scopes tools to users, groups, or caller agents, and can require human approval for sensitive write actions before they run.

Credal builds on outside models. TechCrunch reported in 2023 that Credal does not serve language models itself, and Credal describes a model-agnostic platform that connects to a customer’s own Azure OpenAI deployment, self-hosted models, or frontier models from OpenAI, Anthropic, and others. Its AI work is curating context and routing each workflow to its best-value model, with recommendations for a cheaper model where its accuracy matches.

The Agent Builder includes native evaluation so builders can test an agent before deploying it, and multi-agent workflows keep each step’s own permissions and audit trail. Credal’s Checkr case study reports Salesforce answer accuracy rising from 13% to 86% after Checkr built scoped MCP servers, a result Credal published.

Credal’s go-to-market is enterprise sales led by a pilot. Most customers start with a time-boxed pilot scoped to one team or use case, and Credal says a full enterprise rollout with SSO and custom workflows typically takes a few weeks.

Credal sells on named customers and case studies. It lists Wise, MongoDB, Lattice, Checkr, Comcast, Flatiron Health, IFRS, and Guidewheel among its customers, and Credal describes deployments at Checkr and Guidewheel in its case studies. Its Y Combinator launch post adds the US Federal Government (HHS), and federal spending data shows an HHS-funded contract for a generative AI platform signed in August 2025.

Credal’s scale figures come from its founders. Fischer told TechCrunch in 2023 that Credal had 11 customers. In a February 2025 interview, Ravin told Frontlines that Credal was growing 20% to 30% a month. Credal does not disclose current revenue or customer count.

Credal sells an Enterprise plan with custom pricing built around a customer’s teams, data, and usage. Credal charges for builder seats for people who create agents, MCP servers, and guardrails, plus seats for agent users. It also charges for data indexing by connected data volume and model tokens by usage.

Every plan includes SSO, SAML, and SCIM, SOC 2 Type II, GDPR, and CCPA coverage, consolidated audit logs with SIEM export, and a dedicated support channel. Because the bill combines seats with indexed data and model tokens, a customer’s cost depends on headcount, connected data, and AI consumption together. Credal’s cost attribution shows which team, workflow, and model drives the token portion.

Credal offers four deployment options. A customer can run it as multi-tenant cloud, as a managed single-tenant instance with ingress and egress controls, as a Credal-managed instance inside the customer’s own AWS account or Kubernetes cluster, or air-gapped on its own Kubernetes with optional Credal support as contractors.

Credal keeps model access in the customer’s tenant as well. Bring-your-own-key keeps AI traffic in the customer’s tenant. Customers can access Anthropic models in their VPC through AWS Bedrock or GCP Vertex, and OpenAI models through Azure OpenAI. Credal describes zero-data-retention or equivalent agreements with OpenAI, Anthropic, and Cohere.

Implementation is short by Credal’s account. Most teams are running within days, a full rollout typically takes a few weeks, and Credal’s onboarding team supports customers throughout pilots.

Credal builds its trust case for regulated buyers on attestations and deployment control. It states SOC 2 Type II certification, HIPAA-compliant configurations, GDPR-ready data processing agreements, and participation in the EU-US Data Privacy Framework, its UK Extension, and the Swiss-U.S. Data Privacy Framework, and it points security teams to a trust portal for documentation.

Credal also describes its own security engineering. Its security page says it runs SAST in continuous integration, mirrors source-system permissions for every user, and can detect and redact or block sensitive data before it leaves Credal or the customer’s VPC. Self-hosted and in-VPC options let a buyer keep its data and model traffic in its own environment.

Credal positions itself as one governed layer across AI tools and models. It publishes agents and MCP servers into Claude, ChatGPT, Cursor, and Slack, registers first-party and third-party MCP servers, and offers more than 1,000 connectors with permissions synced from source systems.

Platforms that govern agents themselves compete for the same role. Microsoft says Agent 365 governs agents built on Microsoft, open-source, or third-party platforms. Glean documents an MCP server that exposes its permission-aware context to ChatGPT, Claude, and Copilot Studio, so a customer can get parts of the same governance from those vendors.

Credal’s founders both come from Palantir. TechCrunch reported that Jack Fischer and Ravin Thambapillai both worked at Palantir before founding Credal. Credal’s team page describes Ravin’s seven years building AI systems for financial, life-sciences, and public-health institutions and Jack’s tech-lead work across air travel, healthcare, and US defense.

Ravin told Frontlines that the founders pivoted from an AI chief-of-staff tool to AI security and that Checkr paid for the security product before any other customer. Credal describes a small team drawn from Palantir, Google, and NASA, and Y Combinator’s profile lists 14 people.