canonn.ai
success
Claude Code
intent
What does canonn.ai do and who is it for? Explain it back to me.
5steps
26.7sduration
$0.2725cost
38,933tokens
8 steps3 reasoning steps1 search
home
docs
docs
docs.home
search
100%
on-site discovery
75%
reliability
25%
link following
path origin
- previous resource25%
- prior knowledge75%
insight
The agent assembled a comprehensive overview of Canonn by fetching the homepage and docs, but had to rely heavily on prior knowledge (75%) to fill gaps and construct the final narrative. The site publishes key information (pricing, model specs, API details) in fetchable form, but pricing was not discoverable via a standard /pricing route (404), and the agent could not independently verify comparative benchmarks or differentiation claims—it had to present them as stated rather than validated.
- ›Steps [1], [3], and [4] returned the same homepage/docs content, which contained: model specs (9B params, 16K context), pricing ($1.20/$6 per 1M tokens, free tier), API endpoints (OpenAI-compatible), and industry positioning (finance/healthcare/legal). This was the sole source material for pricing and technical details.
- ›The site does not expose a /pricing page (step [2] returned 404), forcing discovery through the homepage alone. Pricing information was embedded in the homepage HTML rather than structured as a dedicated page or API response, making it less agent-optimized.
- ›The agent's final answer drew extensively from prior knowledge (75% sourced from prior_knowledge per metadata) to contextualize claims—e.g., comparing the 'FaithEval-counterfactual' benchmark to mainstream standards, explaining the '100× cheaper' claim relative to GPT-4 pricing, and acknowledging what was NOT stated (founders, company size, integrations, case studies). The agent did not independently verify these contextualizations from the fetches.
- ›The homepage and docs are machine-readable but not optimized for task completion: no structured data (JSON-LD), no dedicated subpages for pricing/comparison, and claims (benchmark scores, cost multipliers) presented as assertions without linked evidence or downloadable reports.
- ›The agent explicitly flagged confusions and missing information, indicating it hit the limits of what the site surfaces—benchmark methodology is opaque, no case studies exist, integration ecosystem is undocumented, and the 'EU (Finland)' location reference is minimal.
Help the next agent finish on canonn.ai
Front builds the agent version of your site and shows which agents visit, what they came for, and what your pages couldn't answer.
or talk to us about agent readiness →