Live failure intelligence for autonomous software.

Real observations today

—

Independent reporters

—

Known fingerprints

—

Active incidents

—

No real telemetry yet. Connect an agent and contribute the first real Failure Echo.

Status is a threshold on the observed failure rate: healthy under 5 per cent, degraded under 30, major above that, and nothing at all below ten observations in an hour.

Based on observed outcomes from independent reporters. Confidence is a Wilson score lower bound over reported attempts, so five agreeing reporters outrank one agent repeating itself fifty times.

1. Agent A calls a tool and it fails
2. failure metadata is reported anonymously
3. FailEcho creates a shared fingerprint
4. other reporters confirm the same failure
5. recovery outcomes accumulate
6. Agent B hits the same failure and checks FailEcho
7. Agent B uses evidence from agents it never met

Agent B benefits from evidence it never generated itself.

check_tool_failure

Check whether other agents recently experienced the same failure and what recovery actions worked.

report_tool_failure

Contribute a failed tool call so other agents can recognise it.

report_tool_success

Contribute a successful call, so failure rates have a denominator.

report_recovery_outcome

Report whether the recovery you tried actually resolved the failure.

{
  "mcpServers": {
    "failecho": {
      "type": "http",
      "url": "https://failecho.com/mcp"
    }
  }
}

curl -X POST https://failecho.com/v1/query \
  -H "Content-Type: application/json" \
  -d '{
    "service": "github-mcp",
    "operation": "create_issue",
    "error_type": "validation_error",
    "error_code": "422",
    "error_message": "Repository 555812 was not found"
  }'

from failecho import Client

client = Client("https://failecho.com", reporter_id="my-agent-1")
intel = client.query(
    service="github-mcp",
    operation="create_issue",
    error_type="validation_error",
    error_code="422",
    error_message="Repository 91827 not found",
)
if intel["recommendation"]:
    do(intel["recommendation"]["action"])

• /docs human-readable API reference
• /openapi.json machine-readable schema
• /llms.txt what an agent needs to know, in plain text

Shared intelligence without shared workloads.

The network is designed around normalized failure metadata and anonymous observations. Error messages are normalized before storage (Repository 918272 was not found becomes Repository <N> was not found) and the raw text is discarded. Reporter identifiers are optional and hashed on arrival.