IntelligenceQuickstart

Quickstart

Get your first content-bearing trace flowing, then confirm Intelligence can read it. This uses Claude Code as the example; other agents follow the same shape in Connect your agent.

1. Get an API key

Create an sk-tan-… key from Authentication and export it:

export TANGLE_API_KEY=sk-tan-...

2. Point your agent at Intelligence, with content on

Content rides the OTLP logs signal and ships off by default, so the logs exporter and the content gates are both required. Telemetry alone gives you a metadata-only feed.

export CLAUDE_CODE_ENABLE_TELEMETRY=1
export OTEL_EXPORTER_OTLP_PROTOCOL=http/json
export OTEL_EXPORTER_OTLP_ENDPOINT=https://intelligence.tangle.tools/v1/otlp
export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer ${TANGLE_API_KEY}"
 
export OTEL_LOGS_EXPORTER=otlp          # required: content rides log events
export OTEL_METRICS_EXPORTER=otlp       # tokens / cost
export OTEL_TRACES_EXPORTER=otlp        # spans
export CLAUDE_CODE_ENHANCED_TELEMETRY_BETA=1   # required for spans
 
export OTEL_LOG_USER_PROMPTS=1          # task/prompt text
export OTEL_LOG_TOOL_DETAILS=1          # tool arguments
 
# group runs so Intelligence rolls them up by project
export OTEL_RESOURCE_ATTRIBUTES=tangle.subject.key=my-project

The full flag set, privacy tiers, and per-agent config for Codex, OpenCode, and sandbox-run agents are in Connect your agent.

3. Run one real task

Run the agent once on an actual task, not an empty prompt. Intelligence reads the prompts, completions, and tool I/O that a real run produces.

claude -p "fix the failing test in src/auth"

4. Verify content is flowing

Within about 30 seconds, ask Intelligence whether content landed. This endpoint checks both spans and log records, so it sees content wherever it rides:

AUTH="Authorization: Bearer ${TANGLE_API_KEY}"
curl -sS "https://intelligence.tangle.tools/v1/onboarding/status" -H "$AUTH" \
  | jq '{tracesSeen, contentSeen, byAgent}'

contentSeen: true means content is flowing. If tracesSeen is true but contentSeen is false, you missed a gate. tangle connect claude-code --verify runs the same check as one command. See Connect your agent.

Next

  • Connect your agent: Codex, OpenCode, and agents that only emit content when run inside a Tangle sandbox.
  • Run your agent in the sandbox runtime and its evidence lands in the same timeline without any exporter setup.
How it worksConnect your agent