RunTrim
AI Agent Control Layer

Stop AI agents from drifting through your codebase.

RunTrim gives coding agents scope, memory, finish checks and restore points, which often reduces repeated context loading and shortens recovery loops.

runtrim · guarded runlive
taskfix checkout bug
memoryloaded
scopeapi/webhooks/**
forbidden.env migrations/**
risklow
finishtests pass
syncready
guardedstep 4 / 7 · writing tests
01AI AGENT DRIFT

AI agents are fast. They also drift.

One prompt can become twelve files, wasted tokens and broken state.

Scope creep

One task becomes twelve files

Small requests drift into unrelated files.

SESSION_NEW · files_touched: 12 / 3
Token burn

The agent rereads the repo

Bad runs spend context before they solve anything.

DELTA · usage: +38% vs scoped
Broken state

Small fixes break working code

Checkout, auth or config can change without warning.

USAGE · risk: hidden
No recovery

Git revert is too blunt

RunTrim rewinds the AI run, not your whole branch.

VIOLATION · restore: local
No proof

The agent says done

RunTrim checks scope, diff and forbidden writes.

FINISH · finish: verified
Context reset

Every session starts cold

RunTrim carries project memory forward.

RUN_LOG · memory: loaded
02THE RUNTRIM PROTOCOL

Start every AI run with a contract.

Task, scope, memory, rules. Then finish checks the diff.

01
$runtrim start

Creates memory, rules and MCP snippets.

02
$runtrim doctor

Checks if your agent is RunTrim-ready.

03
$runtrim agent "task" --copy

Creates a scoped handoff. MCP agents can create it directly.

04
$runtrim finish

Checks scope, changed files and proof gaps.

05
$runtrim ci check

Blocks risky AI-generated PRs before merge.

06
$runtrim restore last --preview

Preview and rewind a broken AI run locally.

Avg run time2m 14sdown 41% vs unguarded
Tokens per task38.2kdown 27% vs unguarded
Finish-check pass rate94.1%first attempt

Early internal benchmark across repeated guarded vs unguarded tasks. Public benchmark report coming soon.

03Run visibility

Every AI run becomes visible.

See the task, contract, changed files, risk, finish verdict, continuation state, and run history in one place.

01

Run history

Every guarded task records the agent, scope, changed files, and outcome.

history every run logged
02

Proof trail

See scope, forbidden files, tests, diff size, and the finish verdict in one place.

finish state visible and auditable
03

Continuation

Next session starts from what actually happened, not from memory loss.

handoff contract-aware
04

Team-ready accountability

Shared state, approvals, and audit logs as your team workflow grows.

governance built in
05

Smaller, focused runs

Scoped context keeps tasks focused and reduces repeated context loading.

context waste lower
06

Fewer out-of-scope edits

Allowed paths and stop rules keep agents inside the contract boundaries.

scope control stricter by default
04Agent compatibility

Works with every agent you already use.

If it can read a prompt or repo instructions, it can run through RunTrim. No SDK, no model lock-in, no rewiring.

connected agents · 0
C
Claude Code
first-class
Cx
OpenAI Codex
first-class
Cu
Cursor
first-class
Gp
ChatGPT
supported
Km
Kimi
supported
Ds
DeepSeek
supported
Oc
OpenClaw
beta
...
Custom agent
any prompt-runner
Bring your own agent.runtrim agent packages task, scope, memory, and constraints into a clean handoff prompt. Paste into any agent UI, or use MCP for compatible agents.
05Plans

Start local. Scale to a team.

The CLI is free forever. Upgrade when you need cloud history, synced memory, recovery metadata and team governance.

Free
$0/ local

The CLI, local memory and recovery for individual AI coding runs.

  • Unlimited local runs
  • Memory, scope and finish checks
  • All supported agents
  • Local run history
  • Local restore for latest run
Pro
$0/ month
3-day trial

Personal agent control with synced memory, cloud history and recovery metadata.

  • Unlimited guarded handoffs
  • Auto-sync dashboard
  • Cloud run history
  • Memory sync
  • Synced restore metadata
  • Savings reports
Builder
$0/ month

Advanced guardrails for founders shipping production code with AI agents daily.

  • Everything in Pro
  • Unlimited projects
  • Proof / drift reports
  • Priority guardrails
  • Multi-project memory
  • Advanced recovery history
Team
From $0/ seat / month

Shared control for teams using AI coding agents.

  • Everything in Builder
  • Shared team state
  • Approvals and audit logs
  • Shared recovery logs
  • GitHub checks and policies (coming soon)

Compare limits, cloud sync, memory, and team controls.

View all plans

FAQ

Quick answers before you install

Does RunTrim upload my source code?

No. Source code stays local by default. Cloud sync stores run metadata only.

Does restore use more agent tokens?

No. Restore is local recovery. You can recover without spending another agent run.

What happens when a run is BLOCKED?

BLOCKED means needs review. The run is recorded as evidence, but is not trusted yet.

Do I need MCP to use RunTrim?

No. MCP is optional. Copy mode works with any agent UI.

What is included in Free vs Pro vs Builder?

Free is local-first for one project. Pro adds cloud sync and history. Builder adds unlimited projects and CI Gate.

Does RunTrim replace Git?

No. Git remains your version history. RunTrim adds control, verification and recovery for agent runs.

Before any AI agent touches your code, run it through RunTrim.

One command installs the protocol. Every run after that is scoped, memorable, and on the record.

$npm install -g runtrim