agentis
understand your ai agents.
Most teams operate ai agents blind. The agent runs in production, something goes wrong, and the only evidence is a flood of unstructured logs — you can't tell what the agent actually did, which tool failed, or why the run went sideways. agentis turns that noise into a story: every run becomes a readable snapshot with a one-line summary, and the full execution path is traced step by step so you can see exactly where it branched, retried, or broke.
Point your agent's logs at a single ingestion url and snapshots start showing up — no sdk lock-in, generic http works with any agent or framework. Read the snapshot, follow the execution path, and when a run fails, ask the built-in llm to debug it: it reads the trace, names the root cause, and suggests the fix. Free to watch one agent; pro unlocks unlimited agents and llm debugging.
classified ticket #4821 as billing, drafted a refund reply, escalated to a human.
billing.getInvoices timed out after 3 retries — could not load the customer's invoices.
answered a password-reset question from the knowledge base, no escalation needed.
looped 6 times re-asking the same clarifying question before giving up.
routed an enterprise complaint to the priority queue and notified the on-call lead.
classified ticket #4825 as a billing dispute and loaded the customer, but billing.getInvoices timed out after 3 retries. fell back to a holding reply and escalated.
not a bug in the agent's reasoning — billing.getInvoices was unreachable (ETIMEDOUT). the agent handled it right by falling back. lower the per-call timeout so it fails fast instead of burning 30s.
how it works
- 01
register an agent
Create an agent in agentis and grab its api key — one key per agent you want to watch.
- 02
point your logs at one url
Send your agent's logs to a single ingestion endpoint over plain http. No sdk, works with any agent or framework.
- 03
read, trace, debug
Read the snapshot, trace the execution path step by step, and ask the llm to debug any run that failed.
a look inside
a few of the screens you'll actually use.
- condenses logs into readable snapshots
- traces every execution path
- ask an llm to debug a run
classified ticket #4821 as billing, drafted a refund reply, escalated to a human.
billing.getInvoices timed out after 3 retries — could not load the customer's invoices.
answered a password-reset question from the knowledge base, no escalation needed.
looped 6 times re-asking the same clarifying question before giving up.
routed an enterprise complaint to the priority queue and notified the on-call lead.
classified ticket #4825 as a billing dispute and loaded the customer, but billing.getInvoices timed out after 3 retries. fell back to a holding reply and escalated.
not a bug in the agent's reasoning — billing.getInvoices was unreachable (ETIMEDOUT). the agent handled it right by falling back. lower the per-call timeout so it fails fast instead of burning 30s.
agentis guides
Ways to use agentis, and how it compares.
- how toHow to debug LLM agents without guessingA practical method for debugging LLM agents: reproduce the run, read the execution path, isolate the first failing step, inspect its inputs and outputs, then fix and re-run.
- use caseAI agent observability explained: why it matters for MLOpsWhat AI agent observability is, why agents are uniquely hard to observe (non-determinism, tool calls, loops), and what to capture to actually debug and trust them in production.
- how toHow to trace an AI agent's execution pathWhat an execution path is, how to instrument your agent to capture each step, and how to read the resulting trace to understand exactly what your agent did and why.
- comparisonagentis vs traditional logging for AI agentsTraditional logging captures lines; agentis captures runs. A fair comparison of how each handles reading the run, loops, LLM-call visibility, debugging, and setup for AI agents.
- use caseUnderstanding AI agent context: a guide for prompt engineersWhat 'context' really means for an AI agent — prompts, memory, tool outputs, and state — and why capturing the full context of each step is essential to debugging agent behavior.
- how toHow to add observability to an OpenAI or Anthropic agentAdd observability to an OpenAI or Anthropic agent with a generic HTTP approach: register the agent, get an API key, post structured logs to one endpoint, and view the runs as snapshots.