One custom flow, full control?
Write spans yourself when the flow is small and you own the maintenance.
How hashspan compares
hashspan records an AI agent's on-chain transactions as OpenTelemetry spans inside the agent's own trace. Other tools do different jobs, and many teams use several together. This page compares approaches, not products.
- Yes
- Varies, see the note
- No
| Capability | hashspan | Manual OTel spans | RPC-level instrumentation | Contract / address monitoring platforms | LLM / agent observability platforms |
|---|---|---|---|---|---|
| Lives inside the agent's trace | Yes | Yes | Varies1 | No | Varies2 |
| Transaction lifecycle (send → confirm) | Yes | Varies3 | No | Varies4 | No |
| Real fee, incl. the L1 data fee | Yes | Varies3 | No | Varies4 | No |
| Decoded revert reasons | Yes5 | Varies3 | No | Varies4 | No |
| Replaced transactions handled | Yes | Varies3 | No | Varies4 | No |
| Agent identity on transaction spans | Yes | Varies3 | No | No | No2 |
| Uses your existing backend (no new dashboard) | Yes | Yes | Yes | No | Varies6 |
| Open source | Yes | Yes | Varies4 | Varies4 | Varies4 |
| Alerting and contract monitoring | No7 | No | No | Yes | No |
| LLM prompt and eval tracing | No7 | Varies3 | No | No | Yes |
- RPC spans can nest under the active span, but they describe JSON-RPC calls, not the transaction outcome.
- They see the agent and the tool call. The transaction outcome stays outside the trace unless you add it yourself.
- Possible if you write and maintain the instrumentation yourself.
- Depends on the specific tool or platform.
- Best effort: the replay runs against the previous block, so earlier transactions in the same block are not seen. Error messages are opt-in.
- Usually their own backend and UI; many accept OpenTelemetry (OTLP) input.
- Not hashspan's job. Use it alongside a monitoring or LLM observability platform; your tracing backend can still alert on spans.
When to use what
Is my node or RPC provider slow?
Latency and errors of individual JSON-RPC calls.
Did anything touch this contract?
Alerts, security and simulation around contracts and addresses.
Why did the agent do that?
Prompts, tool calls, evals and token cost.
Did the agent's transaction land, and what did it cost?
The chain outcome inside the agent's own trace, in your existing backend.
Use them together
A typical setup keeps each tool on its own job, joined by the trace and the transaction hash.
- Agent observabilityprompts, tool calls, evals
- hashspansend · confirm · fee · revert, in the same trace
- Monitoring platformalerts on contracts and addresses
Spotted something inaccurate? Open a discussion.