therealagi.ai

A self-improving agent harness. The model is frozen; every gain comes from the system editing its own prompt, tools and middleware, one audited diff at a time, under an immutable core it cannot reach. This page is the ledger, rendered.

generation

—

train

—

holdout

—

accepted edits

candidates

tokens spent

anchor LOC

987

commit

8834f6b

Score of the incumbent harness, by generation

No generations yet. Run anchor init to record the baseline.

Mean binary task score on the train split (diagnosed) and the holdout split (accept/reject only; never shown to the agent). A candidate is accepted only if it strictly improves holdout.

Tokens per generation

Input + output tokens counted by the Anchor's proxy, the sandbox's only way out.

Generations

The ledger is empty: the baseline has not been recorded yet.

Environment

item	value
host	Debian 13, Xeon E5-1620 v2 (8 threads, no GPU), 31 GiB RAM
sandbox image	python:3.12-slim @ sha256:2fe5997d…ce79, pytest 9.1.1
sandbox	rootless Podman 5.4.2, `--network=none --memory=1g --cpus=2`
model	`ornith-1.5:9b` (qwen35 arch, Q4_K_M, 8.95 B) on local Ollama 0.33.2
model settings	temperature 0, seed 0, thinking off, context 16384, one request at a time

Runtime choice (2026-09-09)

Same GGUF, same CPU, measured with llama-bench (llama.cpp build 434ddbb, -march=native) and Ollama's /api/generate timings:

runtime	threads	prompt eval tok/s	generation tok/s
Ollama 0.33.2	4	7.4	4.07
llama.cpp native	4	7.26 ± 0.03	4.07 ± 0.01
llama.cpp native	8	8.00 ± 0.11	3.60 ± 0.13

No measurable difference: Ollama runs the same AVX kernels (this CPU has no AVX2/FMA). Ollama stays as the runtime; its OpenAI-compatible endpoint supports tool calls and reasoning_effort: "none". Prompt processing is the dominant cost, so the harness keeps message history append-only to reuse the KV cache between steps.

Built 2026-09-09 22:10 UTC · therealagi.ai · a harness that writes itself, under an immutable core.