Replaying a BeeAI ChatModel run with the model server switched off

[xizhuomengcontin]

Verified against the current BeeAI framework today: a ChatModel call can be recorded from outside the process and replayed later with no model server present.

Disclosure: I work on the recording tool (OrcaReplay, Apache-2.0, no paid tier).

The run

$ orca record generic-openai -- python agent.py
beeai said: BEEAI
info recorded run=run_d453efbe6d26 events=6 exit=0

$ taskkill /F /IM ollama.exe          # model server gone
$ orca replay last
beeai said: BEEAI
info replay.done reused=1/1 exact=1 divergences=0 unmatched=0 exit=0

ChatModel.from_name("openai:…") picks the origin up from OPENAI_BASE_URL, so nothing is installed into BeeAI and no code changes — the proxy just sets the variable for the child process.

Two small things from the way in

Both cost me a couple of minutes, so in case they help someone else or point at a docs gap:

• ChatModel.from_name("openai:…") raises at construction if OPENAI_API_KEY is unset, even when you are pointing at a local endpoint that ignores the key entirely. Understandable, but the failure happens before you get anywhere near a request.
• The method is run(), not create(). Some examples I found use create(), and the AttributeError helpfully suggests _create, which is the private one.

What recording is for

Re-running an agent while you change the prompt, the tools or the flow: the model's answers stay fixed, so the difference you see is your change rather than the sampler. Also CI without keys or spend, and chasing one bad response repeatedly for free.

Scope

egress=blocked on replay means model-provider egress. Recorded tool calls still execute for real, so it is not a sandbox. And this captures the HTTP going to the model, not BeeAI's internals — not a substitute for its own instrumentation.