Tool result trimming
Filters boilerplate framing, pagination markers, ANSI escape sequences, repeated headers, and verbose wrappers from CLI, search, file, API, and other tool results before they reach the model.
Edgee Token Compression V2
Edgee Token Compression V2 reduces redundant input context and model output for coding agents through tool-result trimming, task-aware tool-surface reduction, and configurable output brevity. It is designed for teams using Claude Code, Codex, OpenCode, Cursor, and other Edgee-supported workflows.
Token compression for coding-agent workflows
Edgee Token Compression V2 is a token-control feature in Edgee's agent gateway for coding agents. It reduces redundant content sent to and received from language models, rather than summarizing or removing technical information needed for code tasks. The system works across two layers: input compression for context entering the model and output compression for the model's response.
V2 combines improved tool result trimming with two additions: task-aware tool surface reduction and output brevity. It is presented as a drop-in CLI integration for existing coding-agent workflows, including Claude Code, Codex, OpenCode, and Cursor, with no code changes. Edgee reports a 15–20% production token-bill reduction from compression alone across a rolling 30-day customer aggregate; its separate SWE-bench Lite benchmark measured a 50% reduction with all three techniques enabled.
Compression capabilities
Task-aware tool surface reduction
Scores the task against available MCP tools and presents the model with a curated, task-relevant subset instead of the full tool surface on every request. The IDE can still expose all configured tools.
Output brevity controls
Reduces filler, repetitive scaffolding, over-explanation, and markdown overhead in model responses. Users can select light, medium, or hard levels to control aggressiveness.
Two-layer processing
Separates compression into input and output layers. Input processing targets tool results, tool definitions, codebase context, and related context; output processing targets the model's generated response.
Independent technique flags
Provides the three techniques as independently toggleable configuration flags, allowing teams to enable a subset rather than applying every strategy to every API key.
Practical use cases
Long autonomous coding sessions
Apply tool result trimming to shell output, search results, directory listings, file operations, and API responses so long coding sessions carry less redundant context between steps.
MCP-heavy development environments
Use task-aware tool surface reduction when projects expose multiple MCP servers or large tool catalogs. The model receives a smaller, task-relevant tool set without requiring the IDE to disable tools manually.
Output-cost control
Choose a light, medium, or hard output-brevity level when agent responses contain excessive explanation or formatting and the team wants to reduce output tokens while retaining technical content.
Existing agent deployments
Put existing coding-agent subscriptions or API-key workflows behind the Edgee gateway as a drop-in layer, reducing token usage without rewriting the application or agent integration.
Pros and Cons
Pros
- Targets both input context and model output instead of only trimming tool responses.
- Techniques can be enabled independently, allowing teams to choose a more conservative configuration.
- Tool result trimming is designed to remove redundant framing while retaining technical content for code tasks.
- The product is described as a drop-in CLI wrapper with no application code changes and support for existing API keys and plans.
Cons
- The 15–20% production result is an aggregate, not a guaranteed saving; actual results vary with session length, workload, agent terseness, and enabled techniques.
- The 50% SWE-bench Lite result comes from controlled benchmark conditions and should not be treated as a typical fleet forecast.
- The available source does not provide complete installation, authentication, configuration syntax, or detailed operational limits.
FAQ
How is Edgee Token Compression V2 added to a coding-agent workflow?
Edgee describes Compression V2 as a drop-in CLI wrapper that works with existing API keys and plans, with no code changes. The source does not provide full installation or authentication steps, so setup details should be confirmed in the product documentation.
Can the compression techniques be enabled separately?
The three techniques are independently toggleable configuration flags: tool result trimming, tool surface reduction, and output brevity. Output brevity supports light, medium, and hard levels.
What does Edgee compress?
Tool result trimming removes redundant framing from tool results while preserving technical content. Tool surface reduction curates the tool definitions shown to the model based on the task, while the IDE can continue exposing the full tool set. Output brevity reduces verbosity in model responses.
What level of savings should teams expect?
The production figure is a rolling 30-day aggregate across active Edgee customers and reports a 15–20% reduction in token bills from compression alone. The published SWE-bench Lite result is a separate controlled benchmark in which all three techniques reduced one session from 18,420 to 9,210 tokens, or 50%; it is not a forecast for every fleet.
Are compressed requests safe for coding tasks?
Compression is intended for code-oriented tasks and is described as semantically lossless from the model's perspective. The documentation says extremely short prompts compress less, tool-use schemas are passed through untouched, and Edgee may skip compression when in doubt.
Quick Facts
- Category
- Developer Tool
- Product type
- Coding-agent token compression
- Provider
- Edgee
- Supported agents
- Claude Code, Codex, OpenCode, and Cursor
- Compression layers
- Input and output
- Production result
- 15–20% lower token bills from compression alone in a rolling 30-day customer aggregate
Alternatives à Edgee Token Compression V2
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repositories and verifies changes with the real compiler, debugger, sanitizers, and tests before showing a diff. It offers a free tier plus paid plans, with editor connectors and zero-retention handling described in the source.
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads inside Firecracker micro-VMs. It is designed for workflows that need machine-level isolation, private networking between sandboxes, and programmatic control through SDK, CLI, or MCP.
GitBot
GitBot is a local browser-based workspace for creating reusable coding bots powered by Claude Code, Codex, or OpenCode. It helps developers run those agents across repository folders with persistent threads and configurable permissions.
hob
hob is an independent workspace for coding agents that keeps agent sessions, terminals, history, and follow-up work organized around the tools and providers you already use. It is aimed at developers who want local control over routing, history, and workspace structure rather than a bundled model stack.
Manta AI
Manta AI is an autonomous web app testing tool for teams that want to map application behavior, catch regressions, and generate tests without writing scripts or maintaining selectors. It works from a URL and supports plain-English test flows, run results with screenshots, and scheduled or deployment-triggered checks.
SonOf
SonOf connects to your repo and PM tool, audits the codebase and surrounding product context, and turns approved work into shipped tickets with senior engineering review. It is aimed at founders and engineering leaders who need backlog help without hiring a full team immediately.