Internal AI agentsyour security teamwill approve.
Codi lets you build, govern, and audit agents grounded in your own knowledge — wired into Drive, Confluence, and the tools your teams already use. Every action is recorded, scoped, and explainable.
- 1Confirm new ETA with the supplier (Sysco contact for #1182: Lewis K.).
- 2If new ETA pushes past prep window, swap to backup inventory schedule.
- 3File a delay ticket with reason code DL-04 in ServiceNow within 30 minutes.
Built for the security review.
Most AI tools optimize for a slick demo. Codi optimizes for the moment your CISO opens the architecture diagram and the moment an employee actually trusts the answer.
Build agents grounded in your own knowledge.
Wire Drive, Confluence, SharePoint, Notion, and GitHub into a single retrieval graph. Citations on every answer; no hallucinated policy.
Tamper-evident audit on every action.
Each chat, tool call, and source read is recorded with a signed hash chain. Stream events to your SIEM and retain for 24 months.
Drop agents into the apps you already ship.
REST API, webhooks, and signed events let you call your agents from Slack, Zendesk, or your own SDK. Same audit, same scoping — no separate pipeline to operate.
Guardrails that redact and block — by default.
PII is redacted before it reaches the model. Out-of-policy answers are blocked at runtime and recorded in audit. Configurable per agent, on for every conversation.
Data residency you can put on a contract.
Deploy in us-east, eu-west, or a single-tenant VPC. No customer data leaves the region; no training on your prompts. Ever.
Versioned agents with one-click rollback.
Promote agents through staging → production with diff reviews. Roll back a bad prompt in seconds; every change is signed and attributed.
One platform for the three jobs your team is doing already.
Wire in your sources. Trust the citations.
Connect a Drive folder or a Confluence space in 30 seconds. Codi chunks, embeds, and refreshes on a schedule — and every answer cites the exact paragraph it came from.
- One-click connectors: Drive, Confluence, SharePoint, Notion, GitHub, web
- Per-source ACLs honored — users only see what they're allowed to
- Stale-source alerts when documents drift from the live policy
Every chat. Every tool call. Forever.
The audit log isn't an afterthought — it's the same primitive the agent runs on. Hash-chained, signed, and streamable to your SIEM the same minute an event occurs.
- Signed events · tamper-evident to the byte
- Splunk, Datadog, S3, GCS, and webhook destinations
- PII redaction in transit with a configurable allowlist
| Time | Event | Actor | Result |
|---|---|---|---|
| 10:14:02 | agent.invoke | ms@northwind.com | ok |
| 10:14:03 | kb.retrieve · 3 src | agent:franch | ok |
| 10:14:05 | tool.call · refund | agent:franch | ok |
| 10:14:05 | policy.check | system | ok |
| 10:14:06 | pii.redact · email | system | redacted |
| 10:14:07 | audit.commit | system | ok |
Route by intent. Spend less by default.
Codi inspects each turn and routes to the cheapest model that can answer it. Set hard budgets per agent; cap your worst-case month before it happens.
- Real-time cost meter with projected month-end spend
- Per-agent budgets & per-team chargeback
- Routing recommendations from your own usage
The controls your security team will actually ask about.
SSO & SCIM
SAML 2.0 with Okta, Entra, Google, JumpCloud. SCIM provisioning syncs roles in real time.
Where your data lives
Pin storage and compute to us-east, eu-west, or your private AWS VPC. We don't train on it.
Tamper-evident logs
Signed hash-chained event log. Streams to Splunk, Datadog, S3, and any webhook within 60 seconds.
Scoped permissions
Per-agent ACLs. Per-tool budgets. Approval workflows for dangerous calls. Off by default.
Codi is pre-certification. The controls below are being built into the platform now; audit timelines are committed and we'll share the auditor and start date with prospects under NDA.
From "no LLMs in production" to your first agent in an afternoon.
Connect a source
Auth a Drive folder, Confluence space, or paste a Notion URL. Codi indexes in the background — typically 3–8 minutes for a 1,000-doc workspace.
drive://nw-handbook
● Indexed 412 docs in 4m12s
Define an agent
Pick a model. Pin the sources. Add a system prompt, optional tools, and the team that can use it. Test in the playground; ship to staging when you're ready.
Playground · 20 evals
● 18 of 20 evals passed
Ship to your team
Promote to production with a one-click diff review. Users get it inside the Codi workspace or your own app via the API. Audit is on by default.
Visibility: Store managers
● Live for 1,240 members
Start free. Pay for what you ship.
- 1 agent · 100k tokens
- 24-hour chat history
- Community support
- 1 agent · 1M tokens
- 7-day audit log
- Email support
- 10 agents · 10M tokens
- Custom MCP integrations
- 90-day audit log
- Cost dashboards
- Unlimited agents · BYO keys
- SSO/SCIM · data residency
- 24-month audit · SIEM streaming
- 99.9% SLA
Ship your first agent before the quarter ends.
14-day trial. Bring your own keys, or use ours. A solutions engineer is on standby if your security review needs help.