Ptolemy/Makefile at main · luannn010/Ptolemy · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
APP_NAME=workerd
BIN_DIR=bin

run:
	go run ./cmd/workerd

build:
	mkdir -p $(BIN_DIR)
	go build -o $(BIN_DIR)/workerd ./cmd/workerd
	go build -o $(BIN_DIR)/ptolemy-mcp ./cmd/ptolemy-mcp
	go build -o $(BIN_DIR)/ptolemy ./cmd/ptolemy
	go build -o $(BIN_DIR)/ptolemy-memory ./cmd/ptolemy-memory

test:
	go test -p 1 ./...

test-integration:
	go test -tags=integration ./...

fmt:
	go fmt ./...

tidy:
	go mod tidy

# Phase 0 memory smoke test. Overrides RAG_CHUNK_SIZE_TOKENS for the run so
# small embedding servers (e.g. llama.cpp with --batch-size 64) don't reject
# the input. Everything else (DATABASE_URL, EMBEDDING_*, BRAIN_*) is read
# from .env via cmd/ptolemy's godotenv autoload.
SMOKE_TEST_CHUNK_SIZE ?= 50
SMOKE_TEST_DOC        ?= /tmp/ptolemy-smoke.txt
SMOKE_TEST_DOC_ID     ?= smoke-doc
SMOKE_TEST_QUESTION   ?= What is Ptolemy?

smoke-memory: build
	@if [ ! -f $(SMOKE_TEST_DOC) ]; then \
	  printf "Ptolemy is a Go-based agent runtime project being rebuilt clean-room as v2.\nIt uses a policy harness to gate every side-effecting operation (shellcmd, fileops, gitops, worktrees) behind hybrid approvals: in-band tokens for low-risk commands and out-of-band approval for high-risk ones.\nThe memory module adds Retrieval-Augmented Generation on PostgreSQL with pgvector for dense semantic search.\n" > $(SMOKE_TEST_DOC); \
	fi
	@echo "--- ingest ($(SMOKE_TEST_DOC)) ---"
	RAG_CHUNK_SIZE_TOKENS=$(SMOKE_TEST_CHUNK_SIZE) RAG_CHUNK_OVERLAP_TOKENS=10 \
	  $(BIN_DIR)/ptolemy memory demo ingest $(SMOKE_TEST_DOC_ID) $(SMOKE_TEST_DOC)
	@echo
	@echo "--- ask ($(SMOKE_TEST_QUESTION)) ---"
	RAG_CHUNK_SIZE_TOKENS=$(SMOKE_TEST_CHUNK_SIZE) RAG_CHUNK_OVERLAP_TOKENS=10 \
	  $(BIN_DIR)/ptolemy memory demo ask "$(SMOKE_TEST_QUESTION)"

# Phase 6a capture smoke: runs the REAL BRAIN_* extractor against a sample
# exchange and logs the extracted entries. Requires .env (BRAIN_*).
smoke-capture:
	@set -a; . ./.env; set +a; \
	  go test -p 1 -tags=smoke -run TestExtractorSmoke ./internal/memory/ -v

# Phase 6b consolidation smoke: runs the REAL BRAIN_* LLM through the consolidator
# synthesize step and logs the summary. Requires .env (BRAIN_*).
smoke-consolidate:
	@set -a; . ./.env; set +a; \
	  go test -p 1 -tags=smoke -run TestConsolidatorSmoke ./internal/memory/ -v

# Agent-loop smoke: runs the REAL agent loop end-to-end (NewModule + BRAIN +
# embeddings + eval DB) on a known and an unanswerable seed question. Requires
# .env (BRAIN_*/EMBEDDING_*/DATABASE_URL or MEMORY_EVAL_DATABASE_URL) and the
# fixture corpus.
smoke-agent:
	@set -a; . ./.env; set +a; \
	  RAG_FIXTURE_DIR=eval/testdata/corpus RAG_CHUNK_SIZE_TOKENS=$(EVAL_CHUNK_SIZE) RAG_CHUNK_OVERLAP_TOKENS=10 \
	  AGENT_LOOP_ENABLED=true AGENT_MAX_STEPS=3 \
	  go test -p 1 -tags=smoke -run 'TestAgentLoopSmoke' ./internal/memory/ -v

# Phase 3 memory eval. RAG_FIXTURE_DIR points the binary at the frozen
# fixture corpus under internal/memory/eval/testdata/corpus/; eval.LoadFixtureCorpus
# enumerates the dir and the orchestrator's ingest path chunks/embeds/upserts.
# EVAL_CHUNK_SIZE=20 keeps chunked output under the llama.cpp embedding server's
# 64-token batch ceiling on dense markdown fixtures.
EVAL_SEED         ?= internal/memory/eval/testdata/seed.json
EVAL_FIXTURE_DIR  ?= internal/memory/eval/testdata/corpus
EVAL_CHUNK_SIZE   ?= 20

eval-memory: build
	RAG_FIXTURE_DIR=$(EVAL_FIXTURE_DIR) \
	RAG_CHUNK_SIZE_TOKENS=$(EVAL_CHUNK_SIZE) RAG_CHUNK_OVERLAP_TOKENS=10 \
	  $(BIN_DIR)/ptolemy memory eval -seed $(EVAL_SEED)

eval-memory-agent: build
	RAG_FIXTURE_DIR=$(EVAL_FIXTURE_DIR) \
	RAG_CHUNK_SIZE_TOKENS=$(EVAL_CHUNK_SIZE) RAG_CHUNK_OVERLAP_TOKENS=10 \
	AGENT_LOOP_ENABLED=true \
	  $(BIN_DIR)/ptolemy memory eval -seed $(EVAL_SEED) -agent

eval-memory-sweep: build
	RAG_FIXTURE_DIR=$(EVAL_FIXTURE_DIR) \
	RAG_CHUNK_SIZE_TOKENS=$(EVAL_CHUNK_SIZE) RAG_CHUNK_OVERLAP_TOKENS=10 \
	  $(BIN_DIR)/ptolemy memory eval -seed $(EVAL_SEED) -sweep

eval-memory-dedup: build
	RAG_FIXTURE_DIR=$(EVAL_FIXTURE_DIR) \
	  $(BIN_DIR)/ptolemy memory eval -seed $(EVAL_SEED) -dedup

# Phase 6b synthesis eval (~12 seed scenarios, growable toward 20-40). Uses the
# dedicated eval DB (MEMORY_EVAL_DATABASE_URL, falling back to DATABASE_URL) at the
# real EMBEDDING_DIM. Needs .env (BRAIN_*, EMBEDDING_*).
eval-synth: build
	@set -a; . ./.env; set +a; \
	  $(BIN_DIR)/ptolemy memory synth-eval -scenarios internal/memory/eval/testdata/synth_scenarios.json