Metadata-Version: 2.4
Name: contextspy
Version: 0.1.3
Summary: LLM proxy that analyses token usage in context windows
Author-email: Rimantas Zukaitis <rimasz@gmail.com>
License:                                  Apache License
                                   Version 2.0, January 2004
                                http://www.apache.org/licenses/
        
           TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
        
           1. Definitions.
        
              "License" shall mean the terms and conditions for use, reproduction,
              and distribution as defined by Sections 1 through 9 of this document.
        
              "Licensor" shall mean the copyright owner or entity authorized by
              the copyright owner that is granting the License.
        
              "Legal Entity" shall mean the union of the acting entity and all
              other entities that control, are controlled by, or are under common
              control with that entity. For the purposes of this definition,
              "control" means (i) the power, direct or indirect, to cause the
              direction or management of such entity, whether by contract or
              otherwise, or (ii) ownership of fifty percent (50%) or more of the
              outstanding shares, or (iii) beneficial ownership of such entity.
        
              "You" (or "Your") shall mean an individual or Legal Entity
              exercising permissions granted by this License.
        
              "Source" form shall mean the preferred form for making modifications,
              including but not limited to software source code, documentation
              source, and configuration files.
        
              "Object" form shall mean any form resulting from mechanical
              transformation or translation of a Source form, including but
              not limited to compiled object code, generated documentation,
              and conversions to other media types.
        
              "Work" shall mean the work of authorship made available under
              the License, as indicated by a copyright notice that is included in
              or attached to the work (an example is provided in the Appendix below).
        
              "Derivative Works" shall mean any work, whether in Source or Object
              form, that is based on (or derived from) the Work and for which the
              editorial revisions, annotations, elaborations, or other modifications
              represent, as a whole, an original work of authorship. For the purposes
              of this License, Derivative Works shall not include works that remain
              separable from, or merely link (or bind by name) to the interfaces of,
              the Work and Derivative Works thereof.
        
              "Contribution" shall mean, as submitted to the Licensor for inclusion
              in the Work by the copyright owner or by an individual or Legal Entity
              authorized to submit on behalf of the copyright owner, any work of
              authorship, including the original version of the Work and any
              modifications or additions to that Work or Derivative Works of the Work.
        
              "Contributor" shall mean Licensor and any Legal Entity on behalf of
              whom a Contribution has been received by the Licensor and included
              within the Work.
        
           2. Grant of Copyright License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              copyright license to reproduce, prepare Derivative Works of,
              publicly display, publicly perform, sublicense, and distribute the
              Work and such Derivative Works in Source or Object form.
        
           3. Grant of Patent License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              (except as stated in this section) patent license to make, have made,
              use, offer to sell, sell, import, and otherwise transfer the Work,
              where such license applies only to those patent claims licensable
              by such Contributor that are necessarily infringed by their
              Contribution(s) alone or by the combination of their Contribution(s)
              with the Work to which such Contribution(s) was submitted. If You
              institute patent litigation against any entity (including a cross-claim
              or counterclaim in a lawsuit) alleging that the Work or any
              Contribution embodied within the Work constitutes direct or
              contributory patent infringement, then any patent licenses granted to
              You under this License for that Work shall terminate as of the date
              such litigation is filed.
        
           4. Redistribution. You may reproduce and distribute copies of the
              Work or Derivative Works thereof in any medium, with or without
              modifications, and in Source or Object form, provided that You
              meet the following conditions:
        
              (a) You must give any other recipients of the Work or Derivative
                  Works a copy of this License; and
        
              (b) You must cause any modified files to carry prominent notices
                  stating that You changed the files; and
        
              (c) You must retain, in the Source form of any Derivative Works
                  that You distribute, all copyright, patent, trademark, and
                  attribution notices from the Source form of the Work,
                  excluding those notices that do not pertain to any part of
                  the Derivative Works; and
        
              (d) If the Work includes a "NOTICE" text file as part of its
                  distribution, You must include a readable copy of the
                  attribution notices contained within such NOTICE file, in
                  at least one of the following places: within a NOTICE text
                  file distributed as part of the Derivative Works; within
                  the Source form or documentation, if provided along with the
                  Derivative Works; or, within a display generated by the
                  Derivative Works, if and wherever such third-party notices
                  normally appear. The contents of the NOTICE file are for
                  informational purposes only and do not modify the License.
                  You may add Your own attribution notices within Derivative
                  Works that You distribute, alongside or as an addendum to
                  the NOTICE text from the Work, provided that such additional
                  attribution notices cannot be construed as modifying the License.
        
              You may add Your own license statement for Your modifications and
              may provide additional grant of rights to use, copy, modify, merge,
              publish, distribute, sublicense, and/or sell copies of the
              Contribution, either on an original or modified basis, under the same
              or a different open source license (subject to the conditions of that
              license).
        
           5. Submission of Contributions. Unless You explicitly state otherwise,
              any Contribution intentionally submitted for inclusion in the Work
              by You to the Licensor shall be under the terms and conditions of
              this License, without any additional terms or conditions.
              Notwithstanding the above, nothing herein shall supersede or modify
              the terms of any separate license agreement you may have executed
              with Licensor regarding such Contributions.
        
           6. Trademarks. This License does not grant permission to use the trade
              names, trademarks, service marks, or product names of the Licensor,
              except as required for reasonable and customary use in describing the
              origin of the Work and reproducing the content of the NOTICE file.
        
           7. Disclaimer of Warranty. Unless required by applicable law or
              agreed to in writing, Licensor provides the Work (and each
              Contributor provides its Contributions) on an "AS IS" BASIS,
              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
              implied, including, without limitation, any warranties or conditions
              of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
              PARTICULAR PURPOSE. You are solely responsible for determining the
              appropriateness of using or reproducing the Work and assume any
              risks associated with Your exercise of permissions under this License.
        
           8. Limitation of Liability. In no event and under no legal theory,
              whether in tort (including negligence), contract, or otherwise,
              unless required by applicable law (such as deliberate and grossly
              negligent acts) or agreed to in writing, shall any Contributor be
              liable to You for damages, including any direct, indirect, special,
              incidental, or exemplary damages of any character arising as a
              result of this License or out of the use or inability to use the
              Work (including but not limited to damages for loss of goodwill,
              work stoppage, computer failure or malfunction, or all other
              commercial damages or losses), even if such Contributor has been
              advised of the possibility of such damages.
        
           9. Accepting Warranty or Liability. While redistributing the Work or
              Derivative Works thereof, You may choose to offer, and charge a fee
              for, acceptance of support, warranty, indemnity, or other liability
              obligations and/or rights consistent with this License. However, in
              accepting such obligations, You may offer such obligations only on
              Your own behalf and on Your sole responsibility, not on behalf of
              any other Contributor, and only if You agree to indemnify, defend,
              and hold each Contributor harmless for any liability incurred by,
              or claims asserted against, such Contributor by reason of your
              accepting any warranty or additional liability.
        
           END OF TERMS AND CONDITIONS
        
           APPENDIX: How to apply the Apache License to your work.
        
              To apply the Apache License to your work, attach the following
              boilerplate notice, with the fields enclosed by brackets "[]"
              replaced with your own identifying information. (Don't include
              the brackets!)  The text should be enclosed in the appropriate
              comment syntax for the file format please.  Also, we recommend
              that a file and directory names be included before the boilerplate.
        
              Copyright [yyyy] [name of copyright owner]
        
              Licensed under the Apache License, Version 2.0 (the "License");
              you may not use this file except in compliance with the License.
              You may obtain a copy of the License at
        
                  http://www.apache.org/licenses/LICENSE-2.0
        
              Unless required by applicable law or agreed to in writing, software
              distributed under the License is distributed on an "AS IS" BASIS,
              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
              See the License for the specific language governing permissions and
              limitations under the License.
        
Project-URL: Homepage, https://github.com/RimantasZ/contextspy
Project-URL: Bug Tracker, https://github.com/RimantasZ/contextspy/issues
Keywords: llm,proxy,tokens,observability,mitmproxy,context-window
Classifier: Development Status :: 3 - Alpha
Classifier: Environment :: Console
Classifier: Intended Audience :: Developers
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Topic :: Software Development :: Debuggers
Classifier: Topic :: Internet :: Proxy Servers
Requires-Python: >=3.11
Description-Content-Type: text/markdown
License-File: LICENSE
License-File: NOTICE
Requires-Dist: mitmproxy>=10.0
Requires-Dist: fastapi>=0.110
Requires-Dist: uvicorn[standard]>=0.29
Requires-Dist: sqlalchemy>=2.0
Requires-Dist: tiktoken>=0.7
Requires-Dist: typer>=0.12
Requires-Dist: websockets>=12.0
Requires-Dist: rich>=13.0
Requires-Dist: httpx>=0.27
Requires-Dist: tomli>=2.0; python_version < "3.11"
Provides-Extra: dev
Requires-Dist: pytest>=8; extra == "dev"
Dynamic: license-file

# ContextSpy

A local proxy that intercepts traffic between coding agents (GitHub Copilot, Claude,
opencode, OpenAI SDK scripts, etc.) and LLM APIs — either cloud provider APIs or local
LLM servers — then analyses context composition and token usage.

## Features

- **Two proxy modes**: forward proxy (cloud APIs) and reverse proxy (local LLM servers)
- **Provider support**: OpenAI, Anthropic (Claude), Ollama, llama.cpp, vLLM
- **Agent detection**: Copilot, Claude Desktop, opencode, Cursor, and generic clients
- **Context analysis**: breaks input tokens into 8 categories:
  - System prompt, Tool definitions, Tool results, File contents,
    Conversation history, Current user message, Assistant prefill, Uncategorised
- **Token estimation** via tiktoken (`cl100k_base`)
- **Live dashboard** — real-time WebSocket updates, charts, session grouping
- **Session tracking** — manually start/end named sessions to group requests
- **SQLite storage** — all data stored locally in `~/.contextspy/`

---

## Mode 1: Cloud API Mode (forward proxy)

Use this mode when you want to intercept requests going to **cloud LLM APIs** such as
OpenAI, Anthropic, GitHub Copilot, or Azure OpenAI.

ContextSpy acts as an HTTPS man-in-the-middle proxy. It terminates TLS, inspects the
request, and re-encrypts it before forwarding to the provider. This requires installing
a local CA certificate once so your OS and tools trust ContextSpy's dynamically-signed
server certificates.

### Prerequisites

- Python 3.11+
- [uv](https://github.com/astral-sh/uv) (`pip install uv`) — or plain `pip`
- Administrator / sudo access (for CA cert installation)
- Node.js 18+ and npm — only needed if you want to modify the frontend

### Install

**From PyPI (recommended):**

```bash
pip install contextspy
# or with uv:
uv tool install contextspy
```

**From source:**

```bash
git clone https://github.com/RimantasZ/contextspy.git
cd contextspy
uv venv
uv pip install -e .
```

**Binary releases (macOS / Linux / Windows):**

Pre-built single-file executables are attached to each [GitHub release](https://github.com/RimantasZ/contextspy/releases).
Download and extract the archive for your platform, then run `./contextspy`.

> **macOS Gatekeeper warning** — binaries downloaded from the internet are quarantined
> by macOS and may show a warning that the file "cannot be opened because it was not
> scanned for malware". Remove the quarantine flag after extracting:
>
> ```bash
> xattr -d com.apple.quarantine ./contextspy
> ```
>
> Alternatively, right-click the binary in Finder and choose **Open**, then click
> **Open** in the dialog. This is a one-time step per downloaded binary.
>
> To avoid this entirely, install via **Homebrew** (see below) — Homebrew removes the
> quarantine attribute automatically during install.

**Via Homebrew (macOS — recommended for direct downloads):**

```bash
brew install rimantas/contextspy/contextspy
```

**Via `.deb` package (Ubuntu / Debian):**

Download `contextspy_VERSION_amd64.deb` from the [latest release](https://github.com/RimantasZ/contextspy/releases/latest), then:

```bash
sudo dpkg -i contextspy_*_amd64.deb
```

This installs `contextspy` to `/usr/bin/` and integrates with `apt remove contextspy`.

### Build the UI (optional — only needed if you change the frontend)

The built UI is bundled with the package. Only rebuild if you modify `ui/src/`:

```bash
cd ui
npm install
npm run build   # outputs to contextspy/_web/
cd ..
```

### Step 1 — Start ContextSpy

```bash
# Windows (PowerShell)
uv run contextspy start

# macOS / Linux
uv run contextspy start
```

This starts:
- mitmproxy HTTPS forward proxy on **port 8888**
- FastAPI web dashboard on **port 5173**
- Opens http://127.0.0.1:5173 in your browser automatically

### Step 2 — Install the CA certificate

mitmproxy generates a local CA certificate on first run. You must install it into your
OS trust store so HTTPS connections through the proxy are trusted.

**Automatic install (run once):**

```bash
uv run contextspy install-cert
```

- **Windows**: runs `certutil -addstore Root ~/.mitmproxy/mitmproxy-ca.pem` (requires elevated prompt or UAC prompt)
- **macOS**: runs `security add-trusted-cert -d -r trustRoot -k /Library/Keychains/System.keychain ...` (requires sudo)
- **Linux**: copies cert to `/usr/local/share/ca-certificates/` and runs `update-ca-certificates`

You can also install from the dashboard → **Settings** → **CA Certificate** tab.

> **One-time operation.** The certificate persists in your trust store across restarts.
> You only need to re-run this if you delete `~/.mitmproxy/` or reinstall the OS.

### Step 3 — Configure your agent to use the proxy

Choose the agent you want to intercept and follow the corresponding instructions.
You can also run `uv run contextspy setup-<agent>` for a printed reminder.

#### GitHub Copilot (VS Code)

**Option A — VS Code `settings.json`** (`Ctrl+Shift+P` → "Open User Settings JSON"):

```json
{
  "http.proxy": "http://127.0.0.1:8888",
  "http.proxyStrictSSL": false
}
```

**Option B — environment variables** (set before launching VS Code):

```bash
# PowerShell
$env:HTTPS_PROXY = "http://127.0.0.1:8888"
$env:NODE_EXTRA_CA_CERTS = "$env:USERPROFILE\.mitmproxy\mitmproxy-ca-cert.pem"

# Bash / zsh
export HTTPS_PROXY=http://127.0.0.1:8888
export NODE_EXTRA_CA_CERTS=~/.mitmproxy/mitmproxy-ca-cert.pem
```

Run for the exact snippet:
```bash
uv run contextspy setup-copilot
```

#### Claude CLI / Claude Code

```bash
# PowerShell
$env:HTTPS_PROXY = "http://127.0.0.1:8888"
$env:NODE_EXTRA_CA_CERTS = "$env:USERPROFILE\.mitmproxy\mitmproxy-ca-cert.pem"

# Bash / zsh
export HTTPS_PROXY=http://127.0.0.1:8888
export NODE_EXTRA_CA_CERTS=~/.mitmproxy/mitmproxy-ca-cert.pem
```

> `NODE_EXTRA_CA_CERTS` is required because Claude CLI is an Electron/Node app and has
> its own bundled certificate store that ignores the OS trust store.
> Use `mitmproxy-ca-cert.pem` (cert-only), not `mitmproxy-ca.pem` (key+cert bundle).

Run for the exact snippet:
```bash
uv run contextspy setup-claude
```

#### opencode

```bash
# PowerShell
$env:HTTPS_PROXY = "http://127.0.0.1:8888"
$env:SSL_CERT_FILE = "$env:USERPROFILE\.mitmproxy\mitmproxy-ca-cert.pem"
$env:NODE_EXTRA_CA_CERTS = "$env:USERPROFILE\.mitmproxy\mitmproxy-ca-cert.pem"

# Bash / zsh
export HTTPS_PROXY=http://127.0.0.1:8888
export SSL_CERT_FILE=~/.mitmproxy/mitmproxy-ca-cert.pem
export NODE_EXTRA_CA_CERTS=~/.mitmproxy/mitmproxy-ca-cert.pem
```

> opencode uses both the Go TLS stack (`SSL_CERT_FILE`) and Node.js components
> (`NODE_EXTRA_CA_CERTS`), so both variables are needed.

Run for the exact snippet:
```bash
uv run contextspy setup-opencode
```

#### Python / OpenAI SDK / httpx scripts

```python
import os
os.environ["HTTPS_PROXY"] = "http://127.0.0.1:8888"
```

Or set env vars before running your script:

```bash
# PowerShell
$env:HTTPS_PROXY = "http://127.0.0.1:8888"
python your_script.py

# Bash
HTTPS_PROXY=http://127.0.0.1:8888 python your_script.py
```

#### Generic (curl, httpx CLI, etc.)

```bash
export HTTPS_PROXY=http://127.0.0.1:8888
export HTTP_PROXY=http://127.0.0.1:8888
```

### Step 4 — Use the dashboard

Open http://127.0.0.1:5173. Requests appear in real-time as your agent makes LLM calls.

- **Dashboard** — token usage totals, category breakdown chart, model distribution
- **Requests** — table of all captured requests with token counts and category bars
- **Sessions** — group requests by task; click **Start Session** and give it a name

---

## Mode 2: Local LLM Mode (reverse proxy)

Use this mode when you want to intercept requests going to **local LLM servers** such as
[llama.cpp / llama-server](https://github.com/ggerganov/llama.cpp),
[Ollama](https://ollama.com), or [vLLM](https://github.com/vllm-project/vllm).

**Why a different mode?** When client and server are both on `127.0.0.1`, operating
systems route loopback traffic directly — they bypass `HTTPS_PROXY` entirely. A forward
proxy cannot intercept this traffic. Instead, ContextSpy acts as a **reverse proxy**:
the client connects to ContextSpy's listen port; ContextSpy forwards to the real server.
No TLS, no certificate installation needed.

```
Client (opencode / script)
  base_url = http://127.0.0.1:8889/v1   ← ContextSpy listen port
      │
      ▼
ContextSpy reverse proxy (port 8889)
      │
      ▼
llama-server                            ← actual server
  127.0.0.1:8080
```

### Prerequisites

- Python 3.11+ and uv (same as cloud mode)
- A running local LLM server (llama-server, Ollama, or vLLM) — see setup sections below
- No CA certificate needed

### Step 1 — Configure `~/.contextspy/config.toml`

Add one `[[reverse_targets]]` block per local server you want to intercept. The file is
auto-created at `~/.contextspy/config.toml` on first run (or after running any
`contextspy` command).

```toml
# Example: intercept llama-server running on port 8080
[[reverse_targets]]
name        = "llama-server"
listen_port = 8889
target_url  = "http://127.0.0.1:8080"
provider    = "openai"

# Example: intercept Ollama /v1 endpoint on port 11434
[[reverse_targets]]
name        = "ollama"
listen_port = 8890
target_url  = "http://127.0.0.1:11434"
provider    = "openai"

# Example: intercept vLLM on port 8000
[[reverse_targets]]
name        = "vllm"
listen_port = 8891
target_url  = "http://127.0.0.1:8000"
provider    = "openai"
```

**Fields:**

| Field | Required | Description |
|---|---|---|
| `name` | yes | Human-readable label shown in the CLI and logs |
| `listen_port` | yes | Port ContextSpy binds on `127.0.0.1` |
| `target_url` | yes | Full base URL of the local LLM server |
| `provider` | yes | Response parser: `"openai"` for all three servers above |

All three servers implement the OpenAI-compatible `/v1/chat/completions` API, so
`provider = "openai"` is correct in all cases.

You can run `uv run contextspy setup-llamaserver` (or `-ollama`, `-vllm`) for a
ready-to-paste config snippet and client configuration instructions.

### Step 2 — Start ContextSpy in local mode

```bash
uv run contextspy start-local
```

This starts:
- One mitmproxy reverse-proxy listener per `[[reverse_targets]]` entry
- FastAPI web dashboard on **port 5173**
- Opens http://127.0.0.1:5173 in your browser automatically

If `[[reverse_targets]]` is empty or missing, the command prints a config example and
exits without starting anything.

### Step 3 — Point your client at ContextSpy

Instead of connecting to the LLM server directly, configure your client to use the
ContextSpy listen port.

#### llama-server (llama.cpp)

Default setup: llama-server on port 8080, ContextSpy on port 8889.

```bash
# Print the full setup reminder
uv run contextspy setup-llamaserver

# In your Python / openai SDK client:
from openai import OpenAI
client = OpenAI(
    base_url="http://127.0.0.1:8889/v1",   # ContextSpy port
    api_key="not-needed",
)
```

```bash
# For opencode: set the model's base URL to ContextSpy
# In your opencode config (providers section):
# base_url: http://127.0.0.1:8889/v1
```

#### Ollama

Default setup: Ollama on port 11434, ContextSpy on port 8890.

> Ollama ≥ 0.1.24 is required for the `/v1/chat/completions` OpenAI-compatible
> endpoint. Older versions only have `/api/chat`.

```bash
# Print the full setup reminder
uv run contextspy setup-ollama

# In your Python / openai SDK client:
from openai import OpenAI
client = OpenAI(
    base_url="http://127.0.0.1:8890/v1",   # ContextSpy port
    api_key="ollama",                        # Ollama ignores the key
)
```

> **Alternative for cloud mode:** If your Ollama client respects `HTTPS_PROXY`, you can
> use `contextspy start` (forward proxy mode) instead — Ollama is in the built-in
> hostname filter for `localhost:11434`. Run `uv run contextspy setup-ollama` for both
> options.

#### vLLM

Default setup: vLLM on port 8000, ContextSpy on port 8891.

```bash
# Print the full setup reminder
uv run contextspy setup-vllm

# In your Python / openai SDK client:
from openai import OpenAI
client = OpenAI(
    base_url="http://127.0.0.1:8891/v1",   # ContextSpy port
    api_key="not-needed",
)
```

### Step 4 — Use the dashboard

Same as cloud mode — open http://127.0.0.1:5173. All captured requests appear in
real-time regardless of which local server they came from.

---

## CLI reference

```
contextspy start [--proxy-port 8888] [--web-port 5173] [--no-browser]
    Start in cloud/forward-proxy mode. Opens browser on startup.

contextspy start-local [--web-port 5173] [--no-browser]
    Start in local/reverse-proxy mode. Reads [[reverse_targets]] from config.toml.

contextspy install-cert
    Install mitmproxy CA certificate into OS trust store (cloud mode only).

contextspy status
    Show proxy running state, active session, DB path.

contextspy reset-db [--yes]
    Delete ALL requests and sessions (prompts for confirmation).

contextspy db-stats
    Print database row counts (works offline).

contextspy report
    Print aggregate token stats and category breakdown table (works offline).

contextspy setup-claude        Print proxy env-var commands for Claude CLI
contextspy setup-copilot       Print proxy settings for VS Code / GitHub Copilot
contextspy setup-opencode      Print proxy env-var commands for opencode
contextspy setup-llamaserver   Print config.toml snippet and client URL for llama-server
contextspy setup-ollama        Print config.toml snippet and client URL for Ollama
contextspy setup-vllm          Print config.toml snippet and client URL for vLLM

contextspy session start <name>   Start a named capture session
contextspy session end            End the active session
contextspy session list           List all sessions
```

---

## Architecture

### Cloud mode

```
coding agent → HTTPS_PROXY → mitmproxy (port 8888)
                                  │ TLS terminate + forward
                              cloud LLM API
                                  │
                            ContextSpyAddon
                              → parse, classify, count tokens
                              → write to SQLite
                              → broadcast WebSocket
```

### Local mode

```
client (base_url=:8889) → mitmproxy reverse proxy (port 8889)
                                  │ plain HTTP forward
                            llama-server (port 8080)
                                  │
                            ContextSpyAddon (provider_override="openai")
                              → parse, classify, count tokens
                              → write to SQLite
                              → broadcast WebSocket
```

Both modes share the same FastAPI web server (port 5173), SQLite database, and dashboard.

---

## Data storage

All data is stored in `~/.contextspy/`:

| Path | Description |
|------|-------------|
| `~/.contextspy/contextspy.db` | SQLite database |
| `~/.contextspy/config.toml` | Configuration (auto-created) |

Raw request/response bodies are stored per-request and purged automatically
7 days after capture to save disk space (on next server startup).

---

## Token estimation accuracy

Token counts are **estimates** using tiktoken `cl100k_base` encoding.
Accuracy varies by provider:

| Provider | Expected error |
|----------|----------------|
| OpenAI (GPT-4, GPT-4o) | ~2–5% |
| Anthropic (Claude) | ~5–15% |
| Ollama / llama.cpp / vLLM | ~10–20% |

When the provider reports exact token counts in the API response, those are
stored alongside the estimate for comparison on the request detail page.

---

## Development

### Backend

```bash
uv venv
uv pip install -e ".[dev]"
uvicorn contextspy.api.main:create_app --factory --reload --port 5173
```

### Frontend

```bash
cd ui
npm install
npm run dev   # Vite on :5174, proxies /api to :5173
```

## License

Apache 2.0 — see [LICENSE](LICENSE) and [NOTICE](NOTICE).
