Practical overview
When it fits
Use Paritok when a coding-agent workflow repeatedly sends large tool definitions, file reads, tool outputs or conversation history to an upstream LLM, and you can run a proxy between the agent and that model.
Who it is for
- Developers running coding agents who want to reduce repeated input context sent to an LLM.
- Teams or individuals looking to self-host a compression gateway between an agent and its model API.
Documented uses
- Route coding-agent requests through a local proxy.
- Reduce repeated tool-schema overhead in agent requests.
- Compress tool outputs and file reads before forwarding them upstream.
- Summarize older conversation history as a session's context grows.
- Retrieve original content when compressed or filtered material is needed.
Requirements & limits
The project is public and licensed under Apache-2.0. The documented self-hosted quick start uses Python 3.11 or later and Ollama.
- Paritok is middleware between an agent and an LLM, rather than an agent itself.
- The documented local setup serves the proxy on port 8080 and requires it to remain running while the agent uses it.
- Tool-schema filtering is an optional installation extra; its first request downloads an embedding model for local use.
- The README describes original-content retrieval methods for compressed content and filtered tool schemas.
- Token-saving figures are project-reported estimates and can vary with the request and session.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up Paritok as a local proxy
Install and start the local gateway, then point an agent at its local endpoint.
Before you start
- Python 3.11 or later, as indicated by the repository.
- Ollama installed for the documented self-hosted quick start.
- A separate terminal for launching the agent while the proxy remains running.
Steps
Install the gateway with the command below.
pip install "paritok[proxy]"Run
paritok up; the documented command checks Ollama, obtains the model if needed and serves the proxy on port 8080.In the shell used to launch the agent, set
ANTHROPIC_BASE_URL=http://127.0.0.1:8080, then start the agent.
Expected outcome
The agent sends requests through the locally running Paritok proxy.
Guide limitations
- The proxy runs in the foreground and must remain running during the agent session.
- The README says the default q4 model download is about 2.5 GB on first use.
- This describes the documented setup path, not a separately verified execution.
Compare options for AI coding assistance
These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.
| Project | Documented purpose | Platform | Access and requirements |
|---|---|---|---|
| Paritok | Route coding-agent requests through a local proxy. | Self-hosted, Server | The project is public and licensed under Apache-2.0. The documented self-hosted quick start uses Python 3.11 or later and Ollama. |
| Nasiko | Detect supported coding-agent CLIs and inspect their versions and configuration locations. | CLI | The documented discovery command requires no account. Reporting and routing require an active control plane and valid login. The Docker Compose path requires Docker Engine with Compose V2; the CLI requires Rust 1.85 or later. |
| Sandboxd | Prompt a coding agent to build an app in an isolated container and open its preview URL. | Docker, Self-hosted, Server | The self-hosted project is open source under the MIT license. The repository documents a live browser demo with sample data and no installation. Self-hosting requires Docker with the Compose plugin and Git on Linux; macOS via Docker Desktop is best-effort. |
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 14 October 2026
- Original website title
- Paritok-official/paritok-4b-v1 · Official homepage title · retained in its source language