Structural preflight
Checks known input shapes before a model call is considered.
Little Canary 0.3.6 · Python 3.9–3.13 · Apache-2.0 published-release licenseOpen source · inspectable boundary
Experimental prompt-injection sensing
Inspect untrusted input before an agent acts.
Little Canary sends untrusted language through structural checks and a powerless sacrificial model, then reports routing and inspection coverage separately.
$ uvx --from little-canary==0.3.6 little-canary --versionRelease
v0.3.6
Published on PyPI
Runtime
Python 3.9–3.13
Apache-2.0 published-release license
First run
One pinned command
No project files required
Scope
Bounded tool
Not a system-level guarantee
§ 01
The problem
Why this exists
Make the evidence boundary visible.
An authority-bearing agent can receive untrusted language before it is clear whether inspection took place. Little Canary separates routing disposition from inspection coverage so a degraded or unavailable check is not mistaken for a completed clean inspection.
§ 02
What it does
Powerless canary
Uses a sacrificial model with no application tools or authority.
Explicit coverage
Separates routing disposition from whether the enabled inspection layers ran.
§ 05
Related concepts
GLOSSARYBehavioral CanaryingA proposed approach that uses powerless behavioral probes to inspect untrusted input.GLOSSARYInspection coverageWhether the inspection layers required for a decision actually ran.GLOSSARYIndirect prompt injectionUntrusted text influences an agent through retrieved or embedded instructions rather than a direct user request.GLOSSARYSilent AI failure modeA plausible-looking output that hides a consequential error.
The boundary
Useful evidence is not a blanket guarantee.
Little Canary is an inbound risk sensor, not a security guarantee or agent runtime. It does not claim universal detection, formal security, or aggregate accuracy.
Compare it with the other open-source tools →Read the technical note →