contextpress 0.6.0__tar.gz → 0.6.2__tar.gz

{contextpress-0.6.0 → contextpress-0.6.2}/.gitignore

@@ -32,3 +32,7 @@ Thumbs.db
 
 # Secrets (never commit)
 .env
+
+# Draft articles / writing prompts (local only)
+articles/STYLE_PROMPT_taha_azizi.md
+articles/introducing-contextpress.md

{contextpress-0.6.0 → contextpress-0.6.2}/CHANGELOG.md

@@ -4,6 +4,21 @@ All notable changes to `contextpress` are recorded here.
 The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/)
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.6.2] - 2026-07-27
+
+- **Agent-oriented fixtures** — three offline agent threads under `tests/fixtures/chats/`
+  (large tool JSON, repeated log lines, mixed tool call/result trace).
+- **`CompressionStats.summary()`** / **`CompressionResult.summary()`** — one-line human-readable
+  savings report (tokens, stages, optional USD when ``cost_provider`` is set).
+- Example: `examples/agent_json_compress.py`.
+
+## [0.6.1] - 2026-07-26
+
+- **USD on stats** — ``CompressionStats.attach_cost()`` and optional ``cost_provider`` on
+  ``ContextManager`` / ``compress()`` fill ``estimated_input_cost_*_usd`` and
+  ``estimated_cost_saved_usd`` (included in ``to_dict()``).
+- Opt-in only: without ``cost_provider``, cost fields stay ``None``.
+
 ## [0.6.0] - 2026-07-26
 
 - **`structure` stage** — early Tier 1 compaction: JSON minify, whitespace tighten, consecutive log-line dedupe (stdlib only).

{contextpress-0.6.0 → contextpress-0.6.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: contextpress
-Version: 0.6.0
+Version: 0.6.2
 Summary: Deterministic context compression for LLM chat, RAG, and agent pipelines
 Project-URL: Homepage, https://github.com/Taha-azizi/contextpress
 Project-URL: Documentation, https://github.com/Taha-azizi/contextpress#readme
@@ -415,6 +415,27 @@ est = cm.estimate_cost(messages, provider="openai", model="gpt-4o-mini", output_
 print(est.total_cost_usd, est.to_dict())
 ```
 
+**USD on compression stats** (0.6.1+, opt-in):
+
+```python
+cm = ContextManager(type="chat", model="gpt-4o-mini", cost_provider="openai")
+result = cm.compress(messages, token_budget=2000, return_stats=True)
+print(result.stats.estimated_input_cost_before_usd)
+print(result.stats.estimated_input_cost_after_usd)
+print(result.stats.estimated_cost_saved_usd)
+# or attach later: result.stats.attach_cost(provider="anthropic", model="claude-haiku-4-5")
+```
+
+**Readable savings report** (0.6.2+):
+
+```python
+result = cm.compress(messages, token_budget=2000, return_stats=True)
+print(result.summary())
+# contextpress (chat, medium): 12 -> 8 turns, 842 -> 410 tokens (51.3% saved)
+# stages: structure, filler, repetition, budget
+# est. input cost: $0.000126 -> $0.000061 (saved $0.000065)   # when cost_provider set
+```
+
 See [`ROADMAP.md`](ROADMAP.md) for positioning vs heavier compression stacks and the 0.6.x plan.
 
 ## Tier 1 vs Tier 2 (classical NLP vs LLM)

{contextpress-0.6.0 → contextpress-0.6.2}/README.md

@@ -174,6 +174,27 @@ est = cm.estimate_cost(messages, provider="openai", model="gpt-4o-mini", output_
 print(est.total_cost_usd, est.to_dict())
 ```
 
+**USD on compression stats** (0.6.1+, opt-in):
+
+```python
+cm = ContextManager(type="chat", model="gpt-4o-mini", cost_provider="openai")
+result = cm.compress(messages, token_budget=2000, return_stats=True)
+print(result.stats.estimated_input_cost_before_usd)
+print(result.stats.estimated_input_cost_after_usd)
+print(result.stats.estimated_cost_saved_usd)
+# or attach later: result.stats.attach_cost(provider="anthropic", model="claude-haiku-4-5")
+```
+
+**Readable savings report** (0.6.2+):
+
+```python
+result = cm.compress(messages, token_budget=2000, return_stats=True)
+print(result.summary())
+# contextpress (chat, medium): 12 -> 8 turns, 842 -> 410 tokens (51.3% saved)
+# stages: structure, filler, repetition, budget
+# est. input cost: $0.000126 -> $0.000061 (saved $0.000065)   # when cost_provider set
+```
+
 See [`ROADMAP.md`](ROADMAP.md) for positioning vs heavier compression stacks and the 0.6.x plan.
 
 ## Tier 1 vs Tier 2 (classical NLP vs LLM)

{contextpress-0.6.0 → contextpress-0.6.2}/ROADMAP.md

@@ -37,8 +37,8 @@ deterministic Tier‑1 NLP for chat / RAG / agent **message histories**, with op
 
 | Version | Focus |
 |---------|--------|
-| **0.6.0** | `structure` stage + `estimate_cost()` + this roadmap |
-| **0.6.1** | Wire estimated USD into `CompressionStats` / reports |
-| **0.6.2** | Agent-oriented fixtures for JSON/tool payloads; polish |
+| **0.6.0** | `structure` stage + `estimate_cost()` + this roadmap — shipped |
+| **0.6.1** | Wire estimated USD into `CompressionStats` / reports — shipped |
+| **0.6.2** | Agent-oriented fixtures for JSON/tool payloads; `summary()` report — shipped |
 
 Stay classical-NLP-first; keep optional LLM extras optional.

{contextpress-0.6.0 → contextpress-0.6.2}/contextpress/__init__.py

@@ -16,7 +16,7 @@ __all__ = [
     "CompressionResult",
     "CompressionStats",
 ]
-__version__ = "0.6.0"
+__version__ = "0.6.2"
 
 
 def __getattr__(name: str) -> Any:

{contextpress-0.6.0 → contextpress-0.6.2}/contextpress/core.py

@@ -47,6 +47,7 @@ class ContextManager:
         llm_min_input_chars: int = 1500,
         llm_max_summary_tokens: int = 2048,
         llm_mode: str = "replace_all",
+        cost_provider: str | None = None,
     ):
         if type not in PROFILES:
             raise ValueError(f"unknown context type {type!r}")
@@ -62,6 +63,8 @@ class ContextManager:
         self.llm_min_input_chars = int(llm_min_input_chars)
         self.llm_max_summary_tokens = int(llm_max_summary_tokens)
         self.llm_mode = llm_mode
+        # When set, compress(..., return_stats=True) attaches USD fields on stats.
+        self.cost_provider = cost_provider
         self._custom_stages: dict[str, StageConfig] = {}
 
     def estimate_tokens(self, messages: Any, *, model: str | None = None) -> int:
@@ -181,12 +184,15 @@ class ContextManager:
         disable: list[str] | None = None,
         return_stats: bool = False,
         dry_run: bool = False,
+        cost_provider: str | None = None,
     ) -> Any | CompressionResult:
         """Run the pipeline; return value matches input shape (dict list, tuples, strings, etc.).
 
         ``token_budget`` must be a positive int or None. Unknown keys in ``disable`` are ignored.
         With ``return_stats=True``, returns a ``CompressionResult`` with ``messages`` and ``stats``.
         With ``dry_run=True``, runs Tier 1 only (no LLM calls) and returns the original messages.
+        When ``cost_provider`` (or ``self.cost_provider``) is set and stats are returned,
+        ``stats`` includes approximate input USD before/after compression.
         """
         if dry_run:
             return_stats = True
@@ -220,6 +226,9 @@ class ContextManager:
             out = pipeline.run(conv, stats=stats, dry_run=dry_run)
         if stats is not None:
             stats.warnings_emitted = captured
+            prov = cost_provider if cost_provider is not None else self.cost_provider
+            if prov is not None:
+                stats.attach_cost(provider=prov, model=self.model or "gpt-4o-mini")
         if dry_run:
             messages_out = denormalize_output(clone_conversation(conv), ctx)
         else:
@@ -239,6 +248,7 @@ class ContextManager:
         disable: list[str] | None = None,
         return_stats: bool = False,
         dry_run: bool = False,
+        cost_provider: str | None = None,
     ) -> list[Any] | list[CompressionResult]:
         """Run ``compress()`` on each conversation in ``conversations``."""
         if not isinstance(conversations, list):
@@ -252,6 +262,7 @@ class ContextManager:
                 disable=disable,
                 return_stats=return_stats,
                 dry_run=dry_run,
+                cost_provider=cost_provider,
             )
             for messages in conversations
         ]
@@ -266,6 +277,7 @@ class ContextManager:
         disable: list[str] | None = None,
         return_stats: bool = False,
         dry_run: bool = False,
+        cost_provider: str | None = None,
     ) -> Any | CompressionResult:
         """Async wrapper around ``compress()`` (runs in a worker thread)."""
         return await asyncio.to_thread(
@@ -277,6 +289,7 @@ class ContextManager:
             disable=disable,
             return_stats=return_stats,
             dry_run=dry_run,
+            cost_provider=cost_provider,
         )
 
     def set_compression(self, compression: str) -> None:

{contextpress-0.6.0 → contextpress-0.6.2}/contextpress/stats.py

@@ -7,6 +7,7 @@ from typing import Any
 
 import tiktoken
 
+from contextpress.costs import estimate_token_cost
 from contextpress.models import Conversation, Turn
 from contextpress.normalizer import extract_text_for_processing
 
@@ -51,6 +52,11 @@ class CompressionStats:
     token_budget: int | None = None
     dry_run: bool = False
     warnings_emitted: list[str] = field(default_factory=list)
+    # Optional USD estimates (0.6.1+); filled when cost_provider is set or via attach_cost()
+    cost_provider: str | None = None
+    cost_model: str | None = None
+    estimated_input_cost_before_usd: float | None = None
+    estimated_input_cost_after_usd: float | None = None
 
     @property
     def turns_removed(self) -> int:
@@ -66,6 +72,54 @@ class CompressionStats:
             return 0.0
         return round(100.0 * self.tokens_saved / self.tokens_before, 2)
 
+    @property
+    def estimated_cost_saved_usd(self) -> float | None:
+        """Approximate input-USD saved (None if cost was not attached)."""
+        before = self.estimated_input_cost_before_usd
+        after = self.estimated_input_cost_after_usd
+        if before is None or after is None:
+            return None
+        return round(max(0.0, before - after), 8)
+
+    def attach_cost(
+        self,
+        *,
+        provider: str = "openai",
+        model: str | None = "gpt-4o-mini",
+    ) -> CompressionStats:
+        """Fill USD fields from ``tokens_before`` / ``tokens_after``. Returns self."""
+        before = estimate_token_cost(self.tokens_before, provider=provider, model=model)
+        after = estimate_token_cost(self.tokens_after, provider=provider, model=model)
+        self.cost_provider = before.provider
+        self.cost_model = before.model
+        self.estimated_input_cost_before_usd = before.input_cost_usd
+        self.estimated_input_cost_after_usd = after.input_cost_usd
+        return self
+
+    def summary(self) -> str:
+        """One-line human-readable report for logs, notebooks, and demos."""
+        level = self.compression_level or "default"
+        dry = " [dry-run]" if self.dry_run else ""
+        lines = [
+            (
+                f"contextpress ({self.context_type}, {level}){dry}: "
+                f"{self.turns_before} -> {self.turns_after} turns, "
+                f"{self.tokens_before} -> {self.tokens_after} tokens "
+                f"({self.token_savings_pct}% saved)"
+            )
+        ]
+        if self.stages_run:
+            lines.append(f"stages: {', '.join(self.stages_run)}")
+        saved = self.estimated_cost_saved_usd
+        before_usd = self.estimated_input_cost_before_usd
+        after_usd = self.estimated_input_cost_after_usd
+        if saved is not None and before_usd is not None and after_usd is not None:
+            lines.append(
+                "est. input cost: "
+                f"${before_usd:.6f} -> ${after_usd:.6f} (saved ${saved:.6f})"
+            )
+        return "\n".join(lines)
+
     def to_dict(self) -> dict[str, Any]:
         """JSON-serializable snapshot of this run."""
         return {
@@ -86,6 +140,11 @@ class CompressionStats:
             "token_budget": self.token_budget,
             "dry_run": self.dry_run,
             "warnings_emitted": list(self.warnings_emitted),
+            "cost_provider": self.cost_provider,
+            "cost_model": self.cost_model,
+            "estimated_input_cost_before_usd": self.estimated_input_cost_before_usd,
+            "estimated_input_cost_after_usd": self.estimated_input_cost_after_usd,
+            "estimated_cost_saved_usd": self.estimated_cost_saved_usd,
         }
 
 
@@ -96,6 +155,10 @@ class CompressionResult:
     messages: Any
     stats: CompressionStats
 
+    def summary(self) -> str:
+        """Human-readable report for this compression run."""
+        return self.stats.summary()
+
     def to_dict(self, *, include_messages: bool = True) -> dict[str, Any]:
         data = {"stats": self.stats.to_dict()}
         if include_messages:

contextpress-0.6.2/examples/agent_json_compress.py

@@ -0,0 +1,46 @@
+"""Agent JSON compression + readable savings report (0.6.2+)."""
+
+from __future__ import annotations
+
+import json
+
+from contextpress import ContextManager
+
+payload = {
+    "tool": "search_deploys",
+    "service": "api-v2",
+    "environment": "staging",
+    "events": [
+        {
+            "id": f"evt-{i:03d}",
+            "version": f"2.4.{i % 3}",
+            "status": "healthy" if i % 2 else "superseded",
+            "details": {"replicas": 3, "region": "us-east-1"},
+        }
+        for i in range(12)
+    ],
+    "meta": {"query_ms": 38, "truncated": False},
+}
+
+messages = [
+    {"role": "system", "content": "You are a deploy agent with tools."},
+    {"role": "user", "content": "Summarize recent staging deploys for api-v2."},
+    {
+        "role": "assistant",
+        "content": "Fetching deploy history <tool_call> search_deploys(api-v2, staging)",
+    },
+    {"role": "user", "content": "Tool result:\n" + json.dumps(payload, indent=2)},
+    {"role": "assistant", "content": "Staging has multiple recent releases; latest is healthy."},
+]
+
+cm = ContextManager(
+    type="agent",
+    model="gpt-4o-mini",
+    compression="medium",
+    cost_provider="openai",
+)
+result = cm.compress(messages, token_budget=None, return_stats=True)
+
+print(result.summary())
+print()
+print("tool result preview:", result.messages[3]["content"][:120], "...")

{contextpress-0.6.0 → contextpress-0.6.2}/examples/structure_and_cost.py

@@ -18,18 +18,23 @@ messages = [
     {"role": "assistant", "content": "line\nline\nline\nnext"},
 ]
 
-cm = ContextManager(type="agent", model="gpt-4o-mini", compression="medium")
+cm = ContextManager(
+    type="agent",
+    model="gpt-4o-mini",
+    compression="medium",
+    cost_provider="openai",
+)
 before = cm.estimate_tokens(messages)
-before_cost = cm.estimate_cost(messages, provider="openai")
 result = cm.compress(messages, token_budget=None, return_stats=True)
-after_cost = cm.estimate_cost(result.messages, provider="openai")
+stats = result.stats
 
-print("tokens:", before, "->", result.stats.tokens_after)
+print("tokens:", before, "->", stats.tokens_after)
 print(
     "est. input USD:",
-    f"{before_cost.input_cost_usd:.6f}",
+    f"{stats.estimated_input_cost_before_usd:.6f}",
     "->",
-    f"{after_cost.input_cost_usd:.6f}",
+    f"{stats.estimated_input_cost_after_usd:.6f}",
+    f"(saved {stats.estimated_cost_saved_usd:.6f})",
 )
-print("stages:", result.stats.stages_run)
+print("stages:", stats.stages_run)
 print("user content:", result.messages[1]["content"][:80])

{contextpress-0.6.0 → contextpress-0.6.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "contextpress"
-version = "0.6.0"
+version = "0.6.2"
 description = "Deterministic context compression for LLM chat, RAG, and agent pipelines"
 readme = "README.md"
 license = { file = "LICENSE" }

contextpress-0.6.2/tests/fixtures/chats/09_agent_tool_json.json

@@ -0,0 +1,22 @@
+{
+  "id": "09_agent_tool_json",
+  "type": "agent",
+  "source": "synthetic (large pretty-printed tool JSON)",
+  "messages": [
+    {"role": "system", "content": "You are a data agent with search and fetch tools."},
+    {"role": "user", "content": "Find recent deploy events for api-v2 on staging."},
+    {
+      "role": "assistant",
+      "content": "Searching deploy logs <tool_call> search_deploys(service=api-v2, env=staging, limit=20)"
+    },
+    {
+      "role": "user",
+      "content": "Tool result:\n{\n  \"service\": \"api-v2\",\n  \"environment\": \"staging\",\n  \"events\": [\n    {\n      \"id\": \"evt-001\",\n      \"version\": \"2.4.1\",\n      \"status\": \"healthy\",\n      \"timestamp\": \"2026-07-25T14:22:11Z\",\n      \"details\": {\n        \"replicas\": 3,\n        \"region\": \"us-east-1\",\n        \"notes\": \"Rolling update completed without errors\"\n      }\n    },\n    {\n      \"id\": \"evt-002\",\n      \"version\": \"2.4.0\",\n      \"status\": \"superseded\",\n      \"timestamp\": \"2026-07-24T09:10:03Z\",\n      \"details\": {\n        \"replicas\": 3,\n        \"region\": \"us-east-1\",\n        \"notes\": \"Previous stable release\"\n      }\n    },\n    {\n      \"id\": \"evt-003\",\n      \"version\": \"2.3.9\",\n      \"status\": \"rolled_back\",\n      \"timestamp\": \"2026-07-22T18:44:55Z\",\n      \"details\": {\n        \"replicas\": 2,\n        \"region\": \"us-east-1\",\n        \"notes\": \"Rollback after health check failure on pod-2\"\n      }\n    }\n  ],\n  \"meta\": {\n    \"query_ms\": 42,\n    \"truncated\": false,\n    \"source\": \"deploy-index\"\n  }\n}"
+    },
+    {
+      "role": "assistant",
+      "content": "api-v2 staging is on 2.4.1 (healthy). Prior release 2.4.0 was superseded yesterday."
+    },
+    {"role": "user", "content": "Any blockers for prod?"}
+  ]
+}

contextpress-0.6.2/tests/fixtures/chats/10_agent_repeated_logs.json

@@ -0,0 +1,22 @@
+{
+  "id": "10_agent_repeated_logs",
+  "type": "agent",
+  "source": "synthetic (agent trace with repeated log lines)",
+  "messages": [
+    {"role": "system", "content": "You are a debugging agent."},
+    {"role": "user", "content": "Why is checkout failing on staging?"},
+    {
+      "role": "assistant",
+      "content": "Pulling recent error logs <tool_call> fetch_logs(service=checkout, env=staging, level=error)"
+    },
+    {
+      "role": "user",
+      "content": "Tool result:\nERROR checkout: payment gateway timeout after 30s\nERROR checkout: payment gateway timeout after 30s\nERROR checkout: payment gateway timeout after 30s\nERROR checkout: payment gateway timeout after 30s\nERROR checkout: payment gateway timeout after 30s\nERROR checkout: payment gateway timeout after 30s\nERROR checkout: payment gateway timeout after 30s\nERROR checkout: payment gateway timeout after 30s\nWARN checkout: retry attempt 3/3\nWARN checkout: retry attempt 3/3\nWARN checkout: retry attempt 3/3\nINFO checkout: circuit breaker opened\nINFO checkout: circuit breaker opened"
+    },
+    {
+      "role": "assistant",
+      "content": "Looks like repeated gateway timeouts triggered the circuit breaker."
+    },
+    {"role": "user", "content": "Summarize the root cause in one sentence."}
+  ]
+}

contextpress-0.6.2/tests/fixtures/chats/11_agent_mixed.json

@@ -0,0 +1,32 @@
+{
+  "id": "11_agent_mixed",
+  "type": "agent",
+  "source": "synthetic (mixed user, tool call, tool result, follow-up)",
+  "messages": [
+    {"role": "system", "content": "You are an ops agent with deploy and status tools."},
+    {"role": "user", "content": "We've decided on using the new pipeline for api-v2 staging deploy."},
+    {
+      "role": "assistant",
+      "content": "Acknowledged. Checking current status <tool_call> get_deploy_status(api-v2, staging)"
+    },
+    {
+      "role": "user",
+      "content": "Tool result: {\"service\":\"api-v2\",\"env\":\"staging\",\"version\":\"2.4.1\",\"healthy\":true,\"pods\":[{\"name\":\"api-v2-0\",\"ready\":true},{\"name\":\"api-v2-1\",\"ready\":true}]}"
+    },
+    {
+      "role": "assistant",
+      "content": "Staging is healthy on 2.4.1. I can schedule the new pipeline deploy when you confirm."
+    },
+    {"role": "user", "content": "Go ahead and schedule it for tonight."},
+    {
+      "role": "assistant",
+      "content": "Scheduling deploy <tool_call> schedule_deploy(api-v2, staging, pipeline=new, window=tonight)"
+    },
+    {
+      "role": "user",
+      "content": "Tool result:\n{\n  \"scheduled\": true,\n  \"window\": \"2026-07-26T02:00:00Z\",\n  \"pipeline\": \"new\",\n  \"ticket\": \"DEP-4821\"\n}"
+    },
+    {"role": "assistant", "content": "Deploy DEP-4821 is scheduled for tonight via the new pipeline."},
+    {"role": "user", "content": "Thanks. Basically that's all I needed."}
+  ]
+}

{contextpress-0.6.0 → contextpress-0.6.2}/tests/fixtures/chats/README.md

@@ -13,3 +13,6 @@ No network required — used by `tests/test_fixture_chats.py`.
 | `06_rag_chunks.json` | rag_doc | Chunk + query |
 | `07_short_stable.json` | chat | Minimal change |
 | `08_mixed_ack_resolution.json` | chat | Ack vs resolution (AUDIT C3) |
+| `09_agent_tool_json.json` | agent | Large pretty-printed tool JSON |
+| `10_agent_repeated_logs.json` | agent | Repeated log lines in tool output |
+| `11_agent_mixed.json` | agent | Tool call + result + follow-up thread |

{contextpress-0.6.0 → contextpress-0.6.2}/tests/test_fixture_chats.py

@@ -58,7 +58,7 @@ def test_fixture_recommend_preset(path: Path):
 
 
 def test_fixture_catalog_not_empty():
-    assert len(FIXTURES) >= 8
+    assert len(FIXTURES) >= 11
 
 
 def test_c3_ack_may_weaken_resolution_signal():

contextpress-0.6.2/tests/test_v061.py

@@ -0,0 +1,55 @@
+from contextpress import ContextManager
+from contextpress.stats import CompressionStats
+
+
+def test_attach_cost_on_stats():
+    stats = CompressionStats(tokens_before=1_000_000, tokens_after=500_000)
+    stats.attach_cost(provider="openai", model="gpt-4o-mini")
+    assert stats.cost_provider == "openai"
+    assert stats.cost_model == "gpt-4o-mini"
+    assert stats.estimated_input_cost_before_usd == 0.15
+    assert stats.estimated_input_cost_after_usd == 0.075
+    assert stats.estimated_cost_saved_usd == 0.075
+    d = stats.to_dict()
+    assert d["estimated_cost_saved_usd"] == 0.075
+
+
+def test_cost_fields_none_by_default():
+    cm = ContextManager(type="chat")
+    result = cm.compress(
+        [{"role": "user", "content": "hello basically"}],
+        token_budget=None,
+        return_stats=True,
+    )
+    assert result.stats.cost_provider is None
+    assert result.stats.estimated_cost_saved_usd is None
+
+
+def test_compress_attaches_cost_when_provider_set():
+    cm = ContextManager(type="chat", model="gpt-4o-mini", cost_provider="openai")
+    messages = [{"role": "user", "content": "basically " + ("word " * 80)}]
+    result = cm.compress(messages, token_budget=None, return_stats=True)
+    assert result.stats.cost_provider == "openai"
+    assert result.stats.estimated_input_cost_before_usd is not None
+    assert result.stats.estimated_input_cost_after_usd is not None
+    assert result.stats.estimated_cost_saved_usd is not None
+    assert result.stats.estimated_cost_saved_usd >= 0
+
+
+def test_compress_cost_provider_kwarg_overrides_default():
+    cm = ContextManager(type="chat", model="gpt-4o-mini", cost_provider="openai")
+    result = cm.compress(
+        [{"role": "user", "content": "hello " * 40}],
+        token_budget=None,
+        return_stats=True,
+        cost_provider="local",
+    )
+    assert result.stats.cost_provider == "local"
+    assert result.stats.estimated_input_cost_before_usd == 0.0
+
+
+def test_preview_inherits_cost_provider():
+    cm = ContextManager(type="chat", model="gpt-4o-mini", cost_provider="anthropic")
+    preview = cm.preview([{"role": "user", "content": "hello " * 20}], token_budget=500)
+    assert preview.stats.cost_provider == "anthropic"
+    assert preview.stats.estimated_input_cost_before_usd is not None

contextpress-0.6.2/tests/test_v062.py

@@ -0,0 +1,120 @@
+"""0.6.2 — agent fixtures, stats.summary(), example script."""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+from contextpress import ContextManager
+from contextpress.stats import CompressionResult, CompressionStats
+
+FIXTURES_DIR = Path(__file__).parent / "fixtures" / "chats"
+AGENT_FIXTURES = [
+    FIXTURES_DIR / "09_agent_tool_json.json",
+    FIXTURES_DIR / "10_agent_repeated_logs.json",
+    FIXTURES_DIR / "11_agent_mixed.json",
+]
+
+
+def _load(path: Path) -> dict:
+    return json.loads(path.read_text(encoding="utf-8"))
+
+
+def test_summary_basic():
+    stats = CompressionStats(
+        context_type="agent",
+        compression_level="medium",
+        turns_before=8,
+        turns_after=6,
+        tokens_before=1000,
+        tokens_after=500,
+        stages_run=["structure", "filler", "repetition"],
+    )
+    text = stats.summary()
+    assert "agent" in text
+    assert "medium" in text
+    assert "1000 -> 500 tokens" in text
+    assert "50.0% saved" in text
+    assert "structure" in text
+    assert "est. input cost" not in text
+
+
+def test_summary_includes_cost_when_present():
+    stats = CompressionStats(
+        context_type="chat",
+        compression_level="low",
+        turns_before=4,
+        turns_after=4,
+        tokens_before=200,
+        tokens_after=150,
+        stages_run=["structure"],
+        estimated_input_cost_before_usd=0.000030,
+        estimated_input_cost_after_usd=0.0000225,
+    )
+    text = stats.summary()
+    assert "est. input cost:" in text
+    assert "saved $0.000008" in text
+
+
+def test_summary_dry_run_flag():
+    stats = CompressionStats(
+        context_type="chat",
+        dry_run=True,
+        turns_before=2,
+        turns_after=2,
+        tokens_before=50,
+        tokens_after=50,
+    )
+    assert "[dry-run]" in stats.summary()
+
+
+def test_compression_result_summary_delegates():
+    stats = CompressionStats(
+        context_type="agent",
+        turns_before=3,
+        turns_after=3,
+        tokens_before=100,
+        tokens_after=80,
+    )
+    result = CompressionResult(messages=[], stats=stats)
+    assert result.summary() == stats.summary()
+
+
+def test_agent_tool_json_fixture_structure_saves_tokens():
+    data = _load(AGENT_FIXTURES[0])
+    cm = ContextManager(type="agent", compression="medium")
+    result = cm.compress(data["messages"], token_budget=None, return_stats=True)
+    assert "structure" in result.stats.stages_run
+    assert result.stats.tokens_after < result.stats.tokens_before
+    texts = " ".join(str(m.get("content", "")) for m in result.messages)
+    assert "tool_call" in texts or "api-v2" in texts
+
+
+def test_agent_repeated_logs_dedupes_lines():
+    data = _load(AGENT_FIXTURES[1])
+    cm = ContextManager(type="agent", compression="medium")
+    result = cm.compress(data["messages"], token_budget=None, return_stats=True)
+    assert "structure" in result.stats.stages_run
+    tool_turn = next(m for m in result.messages if m.get("role") == "user" and "Tool result" in str(m.get("content", "")))
+    content = str(tool_turn["content"])
+    assert content.count("payment gateway timeout") == 1
+
+
+def test_agent_mixed_preserves_tool_markers():
+    data = _load(AGENT_FIXTURES[2])
+    cm = ContextManager(type="agent", compression="high")
+    result = cm.compress(data["messages"], token_budget=600, return_stats=True)
+    texts = " ".join(str(m.get("content", "")) for m in result.messages)
+    assert "tool_call" in texts or "DEP-4821" in texts or "pipeline" in texts
+    assert result.stats.tokens_after <= result.stats.tokens_before
+
+
+def test_compress_summary_with_cost_provider():
+    cm = ContextManager(type="agent", model="gpt-4o-mini", cost_provider="openai")
+    messages = [
+        {"role": "user", "content": json.dumps({"hits": list(range(50))}, indent=2)},
+    ]
+    result = cm.compress(messages, token_budget=None, return_stats=True)
+    summary = result.summary()
+    assert "est. input cost:" in summary
+    assert result.stats.estimated_cost_saved_usd is not None