The platform
One control plane for every agent, model, and tool.
Cohesor is the single hop between your coding agents and the models they call. It compresses prompts, routes each request to the right-sized model, brokers MCP tools, and gives you keys, budgets, observability and billing — without touching your agent's code.
Compress
Headroom trims prompts & tool output ~50% before the model sees them.
Route
Simple requests go to an economy model; hard ones to a flagship.
Orchestrate
Keys, budgets, MCP tools, metering & billing in one place.
Observability
Every request, accounted for.
A unified activity feed merges the routing decision and compression stats for each request — so you can see exactly where tokens and dollars go, in real time.
- Tokens original vs. compressed, per call.
- Model served, and the counterfactual it saved you from.
- Active agents in the last 24h and economy-routing rate.
/v1/messages
routed → economy · 12,480 → 6,120 tok
/v1/messages
routed → flagship · 41,300 → 22,100 tok
mcp · github_list_repos
tool result compressed 63%
/v1/chat/completions
routed → economy · 8,010 → 3,900 tok
aig_live · ci-bot
$180 / $200 used
aig_live · alex@acme.com
$46 / $150 used
aig_live · staging
$12 / $50 used
Keys & budgets
Hand out keys. Cap the spend.
Issue a virtual aig_… key per agent, teammate or environment. Set a credit limit on each — enforced in under a millisecond on the hot path — and revoke instantly. Keys are shown once and stored only as a hash.
Model catalog
Right-sized models, transparent rates.
Use our curated catalog or pass any model slug you like. Routing picks the tier per request; you always see the real per-token rate.
| Tier | Example model | Input / 1M | Output / 1M |
|---|---|---|---|
| Flagship | nemotron-3-ultra-550b | $3.00 | $15.00 |
| Balanced | cohere-north-mini | $0.90 | $3.60 |
| Economy | nemotron-3-nano-30b | $0.20 | $0.60 |
Representative catalog rates. Power users can pass raw model slugs directly.
See the whole platform in action.
Connect an agent, watch the savings land in your dashboard within minutes.