Context efficiency + intelligent routing

Make every agent turn earn its context and its model.

CutCtx is a pre-release, local-first tool for compressing selected tool outputs, logs, RAG chunks, and files through compatible proxy and MCP paths. An agent can retrieve retained originals when the required retrieval integration is enabled; model routing is optional and separate.

Private beta · Access by request · Host and auth support vary

Illustrative workflowEvaluation active
Incoming contextAgent history + tools
CutCtx decisionCompress · recover · route or retain
Chosen pathEligible route or requested model
Useful context retained Provider + capability gates Routing evidence available

Provider adapters and host paths are catalogued; live acceptance varies by client and auth mode

OpenAI Anthropic Gemini / Vertex Amazon Bedrock Chosen endpoint Claude Code Codex Cursor Cline Aider

Host support depends on the client, authentication mode, provider, and configuration. See the integration limits before relying on a listed path.

Selected host paths

Context tools around compatible workflows

Wrappers and managed routes exist for selected hosts. Whether a request passes through CutCtx depends on the client, authentication mode, provider, and route configuration. A configured URL or healthy proxy alone does not prove that a request was processed.

One complete operating loop

Improve the context and the decision around it.

On a compatible configured path, CutCtx can inspect requests, reduce selected context, and expose local telemetry. Retrieval and routing require their own enabled integrations and remain subject to host-specific limits.

01

Observe

See context, token, and workflow signals entering a compatible agent or application path.

02

Understand

Classify protocol and request context before an optimization decision is made.

03

Compress / Recover

Reduce selected context; retrieve retained originals only when the required integration is enabled.

04

Route or retain

Use opt-in routing only for clearly eligible work; retain the requested model when safety gates do not pass.

05

Forward

Send the focused request through the provider and compatible endpoint your team selected.

06

Measure

Review local token estimates separately from provider usage, cache, routing, and retrieval costs.

07

Govern

Add policy, identity, audit, retention, tenancy, and entitlement controls only when the deployment needs them.

One operating layer, deliberate boundaries

The runtime, decisions, and controls your rollout needs.

Start with a local evaluation. Add coordination and governance only when the workflow and organization call for them.

Core runtime

Context that stays useful.

Observe, compress, recover, proxy, and measure close to the workflow.

Intelligent optimization

Route or retain deliberately.

Use opt-in routing only when complexity, capability, provider, account, and transport conditions are eligible.

preset: codex-gpt54mini-high
Operator layer

Inspect the result.

Review savings, latency, budgets, routing, health, and diagnostics with evidence from your workload.

Developer and agent access

Choose a tested path.

Source includes SDKs, compatible proxy routes, CLI wrappers, MCP, plugins, and IDE setup paths; publication and live acceptance vary.

See integrations
Optional enterprise controls

Confirm beta scope.

Identity, policy, audit, retention, tenancy, and entitlement controls have separate availability and acceptance gates.

Evaluation to evidence

Make the next decision with your own report.

CutCtx is designed for a low-friction technical evaluation. Use a workflow you already trust, then inspect what changed.

01
Install

Add CutCtx locally without replacing your agent or provider.

02
Wrap

Route a supported coding agent or compatible application through the proxy.

03
Run

Use the workflow normally so the evaluation reflects real traffic.

04
Inspect

Review local token estimates and routing evidence before drawing conclusions about total cost or task quality.

Read results in context

Local token reductions are not provider billing counters. Separate compression from provider input/output, cache reads and writes, and routing; include retries, retrieval calls, and task outcomes when evaluating a full workload.

Control stays close

Local-first by default. Flexible when the deployment grows.

Core transformations can run in an environment you control, while model requests continue to the configured provider. Optional telemetry egress, licensing, and account flows have separate destinations and settings.

Customer environment

The core compression workflow does not require hosted prompt analytics; review optional external service paths separately.

Chosen provider

Keep OpenAI, Anthropic, Gemini/Vertex, Bedrock, or the compatible endpoint selected by your team.

Governed scale

Add shared reporting, policy, identity, audit, retention, and deployment support when the organization needs them.

A path for each stage

Private beta access is scoped with the team.

Plan descriptions are indicative; confirm current terms and beta availability with the team.

Builder

Prove the workflow locally.

Core compression, proxy, CLI wrapping, SDKs, MCP, local dashboard, and Docker.

Team

Share evidence and policy.

Shared analytics, savings reports, policy presets, budget controls, and support.

Enterprise

Operate with governance.

Identity-aware administration, RBAC, audit export, retention controls, fleet APIs, and deployment support.

Start with evidence

Run CutCtx on the workflow you already have.

Install locally, wrap an agent, and inspect the result on your own traffic.

Practical resources

Guides for context efficiency

Use these worksheets to plan and evaluate context budgets, caching, proxy paths, and quality checks.

Browse all CutCtx guides