Architecture
Sandboxed per client. Data stays on Anthropic.
How Fidelic deploys agents — and what that means for your data.
How a Fidelic agent reaches your Slack
Templates, customized by an agent, deployed in isolation.
Every Fidelic agent starts as a vetted template — a pre-built role definition with capabilities, a written constitution, and a published list of limits. The Roster is the catalog of these templates. Each one ships before any customer sees it, so when you hire, you are not waiting weeks for formation. You are hiring a known quantity.
When you hire, an automated configuration layer (itself a Fidelic agent) takes your brief from Tessa’s voice intake and customizes the template against it — pinning the agent’s tone to your brand, the people it should brief, the channels it should watch, the cadence it should follow. There is no human formation cycle. The customization is an automated step.
The customized agent then deploys into an isolated Anthropic-managed environment — one Anthropic project per customer, with the agent running on Claude. Your Slack workspace connects to that environment. Your data — messages the agent reads, files it summarizes, context it carries — flows from Slack to Anthropic’s infrastructure and back. It does not pass through or land on Fidelic-controlled servers.
Where your data lives
The questions every buyer asks, answered the same way every time.
These don’t change between calls. They’re written here once, in plain language, so your security team can read what we tell every customer.
Where does my data go when an agent runs?
Your data flows from your Slack workspace to Anthropic’s infrastructure, where the customized Fidelic agent runs on Claude. Anthropic processes the data to generate the agent’s response — drafts, summaries, briefings — which then flows back to Slack.
Does Fidelic see my data?
Not the contents. Fidelic logs operational metadata — token counts, error events, tool-call metadata — for billing, support, and tuning. Message content, file content, and the model’s responses are not stored or read by Fidelic.
Is my deployment isolated from other customers?
Yes. Each customer gets a separate Anthropic project, with its own configuration, memory, and API credentials. There is no shared database, no shared session state, and no cross-tenant data path.
What about logs and metadata?
Fidelic logs when an agent runs and which tools it calls (for example, that a Slack message was sent or a link was fetched), but not the contents of those calls. You can request a log of your agent’s activity at any time. We don’t sell or share this metadata.
What is Anthropic’s compliance posture?
Anthropic publishes its compliance certifications at trust.anthropic.com. Because Fidelic agents run on Anthropic infrastructure, your deployment inherits that posture. Verify the current state directly on Anthropic’s trust page — Fidelic does not separately audit Anthropic.
What happens if Anthropic has an outage?
Your agent stops responding until Anthropic recovers. Fidelic does not have a fallback model or a redundant provider. We monitor Anthropic’s status page and notify customers when a sustained outage affects their agent. Single-vendor dependency is a deliberate trade-off for the deeper trust story.
What we depend on Anthropic for
We are downstream of Anthropic. We name what that means.
Fidelic’s architecture inherits more from Anthropic than from any other vendor. That makes the trade-offs easy to enumerate.
- The Claude model itself — its capabilities, accuracy, safety alignment, and behavior under load.
- Compute and inference infrastructure where every Fidelic agent runs.
- Data residency posture — customer data stays on Anthropic's side, not on Fidelic-controlled servers.
- Published compliance certifications (SOC 2 Type 2, ISO 27001, ISO 42001, HIPAA-eligibility for the API). Verify the current state at trust.anthropic.com.
- Uptime and availability of the underlying API.
- Future model improvements. When Anthropic ships a new base model — or we ship a new agent skill — every Fidelic agent inherits it automatically. Same shape as a SaaS update; no migration cost, no upgrade purchase, no version pinning.
Limits, in writing
What we don’t claim.
We don’t separately audit Anthropic.
We trust Anthropic’s published compliance documentation, and you should verify it directly at trust.anthropic.com. We are a customer of Anthropic, just as you would be — we are not a Fidelic-side auditor of Anthropic’s controls.
Fidelic doesn’t have an independent SOC 2 yet.
The platform inherits Anthropic’s posture. Fidelic’s own organizational SOC 2 is a future commitment, not a current one. If your procurement requires a Fidelic SOC 2 attestation today, we won’t pass that filter.
We are not zero-knowledge.
Operational telemetry — token counts, error events, tool-call metadata — flows through Fidelic for billing, support, and tuning. The contents of your data don’t. If your security model requires zero Fidelic visibility into anything related to your deployment, this isn’t that.
We can’t guarantee Anthropic’s roadmap.
If Anthropic changes API access policies, model availability, pricing, or compliance posture, that propagates to Fidelic deployments. We track the changes and communicate them. We don’t control them.
We don’t run on-premises.
Every Fidelic deployment runs on Anthropic’s managed infrastructure. If your security policy requires on-premises AI — air-gapped, in-VPC, or in your own cloud account — Fidelic isn’t your fit, and that’s our honest answer.
Compliance and verticals
What to know if you’re regulated.
We are not your compliance officer. These are notes from buyers in regulated industries who deploy on Fidelic.
Financial services
Standard practice — confirm your CCO-approved AI usage policy permits Anthropic API processing of the data your agent will read. Review Anthropic’s published certifications (SOC 2, ISO 27001) at trust.anthropic.com. Fidelic doesn’t separately audit Anthropic’s controls.
Healthcare (HIPAA)
Anthropic offers a Business Associate Agreement (BAA) for the API. If your Fidelic deployment will process PHI, confirm that your existing Anthropic BAA covers it before deploying. Fidelic doesn’t sign BAAs directly — the BAA is between you and Anthropic.
Legal services
Privilege concerns are buyer-specific. We recommend engaging your firm’s IT security team on the data-residency model — customer data flows from your Slack to Anthropic and back, with operational metadata visible to Fidelic — before deployment. We’re happy to walk through the architecture in detail.
Where to next
Honest Question
Isn’t this just a chatbot?
An agent acts on triggers; a chatbot waits for prompts. Honest answer on the difference.
Read the question →
Browse the Roster
See agent templates by role
Each Roster entry publishes its constitution, capabilities, and limits before you hire.
Browse the Roster →
Verify directly
Anthropic’s trust page
Compliance certifications, security posture, and current status — from the source, not us.
Open trust.anthropic.com →
Voice intake
Talk to Tessa (30 minutes)
The intake agent walks through your role brief and configures the template.
Start →