Pass-through starts on
Advanced routing is explicit and policy gated. A global force-pass-through path remains available.
Enterprise AI control plane
One policy plane for every model your teams use.
Set model access, token ceilings, and cost limits for every employee and sub-org. Then route eligible work to the best allowed model without replacing the tools people already use.
Claude Code, Codex, OpenCode, and compatible clients in. Claude, GPT, GLM, DeepSeek, and Kimi out.
Inherited from
Engineering / Platform
Employee policy
Maya Chen
Enforced
Monthly cost
$312 / $500
Monthly tokens
7.4M / 12M
Model family
Access
Claude
Allowed
GPT
Limited
GLM
Allowed
DeepSeek
Blocked
Kimi
Allowed
Smart routing3 approved candidates available
Inside policy
01Identity-awarePolicy follows the employee
02Cross-familyOne catalog across providers
03In the request pathLimits are enforced, not observed later
The control gap
Provider consoles govern a provider. Rispn governs the employee.
An employee can move between Claude, GPT, GLM, DeepSeek, and Kimi from several coding tools in the same day. Rispn gives the enterprise one place to decide what that person can use and how much they can spend across the full model estate.
See the policy model →
Without Rispn
Separate boundaries
Provider AClaudeits users + spend
Provider BGPTits projects + spend
Provider COpen modelsanother control surface
Unified policy plane
Employee / sub-orgmodel access + tokens + cost
- Claude
- GPT
- GLM
- DeepSeek
- Kimi
Control and routing
Control first. Intelligence second.
Routing never sits above enterprise policy. Rispn first resolves identity, budget, and allowed models. Only then can the routing layer optimize the request.
Policy-aware routing
The best model is the one that fits the task and the rules.
Rispn evaluates only the candidates an administrator has approved. It can keep the requested model, select a compatible lower-cost model when the evidence is strong, or abstain when quality risk is unclear.
Open the routing framework →Route decisionLow-risk repository follow-up
Illustrative
01Policy3 candidates allowed
02Qualitypositive safe evidence
03Sessionwarm compatible lane
04Routeapproved efficient model
Original objective
Modernize the authentication layer without breaking compatibility
01Map constraintsinterfaces, risks, tests
02Stage workreviewable work packets
03Validate outcomeone final quality gate
Objective and validation criteria stay intact
Large-task intelligence
Decompose the work. Keep one outcome contract.
For rare, high-confidence large tasks, Rispn can present a visible decomposition plan before upstream spend. The goal is not more activity. It is smaller, safer work packets that preserve the original objective, evidence, and final validation.
See how review works →One gateway, native workflows
No new developer interface.
Rispn sits underneath the tools teams already use and preserves their native experience.
Agent surfaces
- Claude Code
- Claude Desktop
- Codex CLI
- Codex Desktop
- OpenCode
Model families
- Claude
- GPT
- GLM
- DeepSeek
- Kimi
Enterprise posture
Transparent by default.
Prompt privacy starts strict
Normal hosted traffic does not retain exact prompt text or snippets by default.
SSO anchors identity
Enterprise identity maps every request to the right employee, sub-org, and inherited policy.
Evidence stays labeled
Provider-reported selected cost is separated from projected routing and decomposition estimates.
Start with control
Put every employee and every model inside one governed path.
Begin with policy enforcement. Add smart routing only where your evidence supports it.