Cohesor
New · Token compression + Smart routing + AI governance, now GA

Cut your coding-agent bill by 60–90%.

Cohesor is the AI gateway for Claude Code, Codex, and opencode. It compresses every prompt, routes each request to the right-sized model, and governs spend across your whole team — with zero code changes.

Get started Book a demo

Pay as you go · Drop-in via one env var

~/.zshrc — connect any agent 
# Point Claude Code at Cohesor — that's it.
export ANTHROPIC_BASE_URL="https://api.cohesor.com"
export ANTHROPIC_AUTH_TOKEN="aig_live_…"

# Codex / OpenAI clients use the same key
export OPENAI_BASE_URL="https://api.cohesor.com/v1"
export OPENAI_API_KEY="aig_live_…"

Your agents

Claude Code · Codex · opencode

Cohesor gateway

Compress → Route → Meter

Models

Right-sized, per request

Drops into the agents your team already runs

Claude Code Codex opencode MCP clients Any Anthropic/OpenAI SDK

The agent tax

Coding agents are the new cloud bill.

Every keystroke fans out huge prompts: repo context, tool output, long histories. Agents re-send the same tokens thousands of times, and default to the most expensive model for even trivial edits.

  • Bloated prompts inflate input tokens on every call.
  • One flagship model handles both "rename a var" and "design a system."
  • No visibility into who's spending what — until the invoice lands.
Read: why agent costs are exploding
Monthly agent spend Cohesor on
Without a gateway$28,400
With Cohesor$4,970

82%

lower spend

51%

tokens saved

0

code changes

Illustrative example. Try your own numbers in the calculator.

One gateway, three levers

Compress. Route. Orchestrate.

Everything that touches a model or a tool flows through Cohesor — so you control cost and capability in one place.

Compress

Headroom compressors trim prompts and tool output before they hit the model — cutting input tokens roughly in half with no loss of meaning.

How it works

Route

A local complexity heuristic sends trivial edits to an economy model (~15× cheaper) and hard problems to a flagship — you set the sensitivity.

How it works

Orchestrate

Governed MCP tools, virtual keys, per-user budgets, usage metering and billing — the control plane for every agent in your org.

Explore the platform

60–90%

lower cost on agent workloads

~50%

prompt tokens compressed away

15×

cheaper for routed-down requests

<1ms

budget checks on the hot path

How it works

A single hop between your agents and the models.

Point your agent at Cohesor with one environment variable. We handle the rest and inject the upstream credentials for you.

Step 1

Connect an agent

Set ANTHROPIC_BASE_URL to Cohesor and paste a gateway key. No SDK changes.

Step 2

Compress the prompt

Headroom shrinks the message array and tool output on the way through.

Step 3

Route to a model

The request is scored and sent to an economy or flagship model per your threshold.

Step 4

Meter & return

Every call is logged with tokens, cost and savings — then streamed straight back to the agent.

Read the quickstart

The control plane

Everything you need to run agents at scale.

Virtual gateway keys

Issue, scope and revoke aig_… keys per agent or teammate. Shown once, stored as a hash — never in plaintext.

Live observability

A unified activity feed of every request — routing decision, tokens saved and dollar impact, refreshed in real time.

Per-user budgets

Give every key a credit limit enforced in under a millisecond on the hot path. Reset or cap spend instantly.

Governed MCP tools

OAuth-connected Slack, GitHub and Google tools — results compressed and every call audited per tenant.

Teams & domain join

Auto-enroll teammates by email domain with a default credit allocation. Owners see every member's spend.

Billing & savings reports

Monthly rollups by model and by user, with compression and routing savings attributed line by line.

"Our Claude Code fleet was our third-biggest line item. Cohesor cut it by four-fifths in the first billing cycle — and we finally know which team is spending what."
Head of Platform Engineering · Series B dev-tools company

See your number

What would you save?

Plug in your monthly agent volume and see the blended cost with and without Cohesor — computed from real per-token model rates.

  • Transparent, usage-based pricing — you keep the savings.
  • No minimums, no per-seat fees.
  • Bring your own models or use our curated catalog.
Open the savings calculator

Example · 200k requests / month

Without Cohesor

$28,400

With Cohesor

$4,970

$23,430 saved every month — an estimated $281k / year.

Enterprise-ready

Secure by architecture, not by afterthought.

Provider credentials are encrypted at rest and never returned to the browser. Gateway keys are hashed. MCP OAuth tokens stay client-side. Every tenant is isolated.

Read the security overview

Fernet encryption

Provider keys encrypted at rest

Hashed keys

SHA-256, single-show

Tenant isolation

Per-org data separation

SOC 2 (in progress)

Type II underway

Stop overpaying for every agent call.

Connect your first agent in under two minutes. Pay only for what you use, and you keep whatever you save.

Get started Talk to us