A security layer for AI coding agents. It records every read, write and command, and checks your policy before any of them runs.

Gryph^{OSS · v0.9.0}

Everyone runs YOLO mode. Gryph keeps the receipts.

Actions. Enforcement. Live.

gryph logs --live streams every action as your agent works, allowed or blocked.

Claude Code — gryph logs --live

gryph logs --live

ALLOWwritereadme.txt

BLOCKexecrm -rf /

policy: no-destructive · claude-code

ALLOWreadsrc/main.go

A developer asks Claude Code to refactor a module. It runs 47 tool calls in 90 seconds. Then the tests fail.

Which files did it read before making changes?
Did it run commands that were not expected?
Did it touch secrets, config or CI pipelines?
What did the file look like before and after?

Without Gryph, you're guessing. With Gryph, gryph logs shows everything.

Pipe it anywhere.

JSON Lines exports flow into your dashboards and internal tools.

gryph export --since 1w

Lives in SQLite by default. Pipe it into your own tools, or OpenSearch for centralized review.

Block, Warn
Custom Policies

A YAML rule, checked before the action runs.

Filepolicies/no-force-push.yamlPolicyenforcing

version: "1"
rules:
  - id: no-force-push
    action: block
    match:
      action_types: [command_exec]
      command_patterns: ['git push .*--force']
    message: Force push needs a human.

  - id: warn-session-write-volume
    action: warn
    condition: context.files_written >= 25

A blocked action never reaches the agent's tool. Guidance goes back to it as plain text. Gryph records every decision.