Antibody

Monitor

  • Overview
  • Traffic
  • Incidents
  • Fixes

Protect

  • Guardrails
  • Agents

Setup

  • Integrations
  • Settings
  • Connect an agent
Checking systems…
RRehaanOn callON CALL

airo/sales-research-agent/Connect an agent

--:--:-- UTC

Connect an agent

Five steps, about 15 minutes. This is exactly how sales-research-agent is wired.

  1. 1

    Trace your agent with neatlogs

    Every run becomes a trace Antibody can read. One line at startup, one span per run.

    pip install neatlogs

import neatlogs
neatlogs.init(api_key=NEATLOGS_API_KEY, workflow_name="company-research")

with neatlogs.trace("company-research"):
    brief = run_agent(domain)
  2. 2

    Score every run

    Deterministic tripwires catch silent failures (an empty answer after a tool error, an invented person, a skipped check). Jev adds a grounding score. The scores travel on the trace.

    from detection.checks.score import score_run

score = score_run(run)          # tripwires + Jev
# violations -> neatlogs span "antibody.score"
  3. 3

    Turn on Antibody's detections

    Creates one neatlogs detection per tripwire. When one keeps firing, Antibody opens an incident.

    uv run antibody neatlogs-setup
uv run antibody watch
  4. 4

    Connect GitHub and Entire

    Antibody opens a pull request for every proven fix, and reads the Entire checkpoint to quote why the breaking change was made.

    gh auth login
entire login
entire repo mirror add /gh/<owner>/<repo>
  5. 5

    Pick who gets called

    Open @anti_body_bot on Telegram and send /oncall. When a fix is proven, Antibody calls that chat; only “approve fix N” merges it.

    /oncall