Compresr Context Gateway
Pre-compute conversation summaries in the background so coding agents spend less time waiting when context fills up.
- Source repo
- Compresr-ai/Context-Gateway
- Stars
- ★ 644
- Last updated
- 1d ago
- License
- Apache-2.0
- Primary language
- Go
- FA score
- 36/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformClaude Code
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Claude Code developers who regularly reach context limits can pre-compute summaries to reduce the wait for compaction.
- Not a fit if
- Users whose agents are not Claude Code, Cursor, OpenClaw, or a custom configuration
- Source review
- 36/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
Context Gateway is a proxy placed between an AI agent and an LLM API, providing background history compaction and context optimization. Run context-gateway to open its terminal UI and choose Claude Code, Cursor, OpenClaw, or a custom agent configuration. The setup wizard can configure a summarizer model and API key, a compression trigger threshold (75% by default), and optional Slack notifications. The README says summaries are pre-computed in the background so compaction does not make users wait when a conversation gets long. It provides an installer and points to logs/history_compaction.jsonl for activity; it does not specify supported operating systems, model providers, or the detailed deployment boundary.
The install command fetches and runs a script from https://compresr.ai/api/install. Starting context-gateway opens a TUI where users select claude_code, cursor, openclaw, or custom, then configure the summarizer model, API key, optional Slack notifications, and compression threshold. The Gateway sits between the agent and LLM API and compacts long conversation histories in the background; the README says summaries are pre-computed. Users can inspect logs/history_compaction.jsonl for compaction activity. The source does not describe the configuration file format, log fields, or the protocol between the agent and API.
- Claude Code developers who regularly reach context limits can pre-compute summaries to reduce the wait for compaction.
- Cursor users working through long coding sessions can have conversation history compacted in the background.
- OpenClaw users who want to configure a summarizer model and a compaction trigger threshold.
- Teams with a custom agent configuration that want a history-compaction layer between the agent and an LLM API.
- Developers who need to inspect compaction activity can check
logs/history_compaction.jsonl.
How do you install or deploy this agent?
The README documents installing by downloading and running a shell script. It does not specify supported operating systems or runtime prerequisites. Users also need an API key for the chosen summarizer model; the provider is not named.
curl -fsSL https://compresr.ai/api/install | shHow do you use this agent?
After installation, run the command to open the interactive TUI, select an agent, and configure it. Settings include the summarizer model and API key, a 75% default compaction threshold, and optional Slack notifications. The README does not provide a non-interactive setup path or a copyable configuration file example.
context-gatewayWhat are this agent's strengths and limitations?
- Compaction runs in the background; the README says summaries are pre-computed to reduce waiting when context fills up.
- The TUI offers setup options for Claude Code, Cursor, OpenClaw, and custom agent configurations.
- Users can set the compaction trigger threshold, which defaults to 75%.
- The README names
logs/history_compaction.jsonlas a place to inspect compaction activity.
- Setup calls for a summarizer model and API key, but the README does not name supported models or providers.
- Installation fetches and executes a script from
compresr.ai, so it requires network access during installation. - The README does not specify supported operating systems, runtime prerequisites, or the protocol between the agent and LLM API.
- Slack notifications are optional, but the required credentials and notification contents are not documented.
How does this agent compare with similar options?
The README describes Context Gateway as a proxy for Claude Code, Cursor, and OpenClaw, but does not compare it with other compaction tools or alternatives.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Compresr Context Gateway This agent | 36 · Major gaps | CLI | ★ 644 | 1d ago | Go | Claude Code |
| NoPUA — Unlocking AI's Potential Through Love | 52 · Major gaps | Agent plugin / skillFree + model costs | ★ 1.4k | 24d ago | Python | Codex · Claude Code |
| Hivemind — Shared Brain for Your Agents | 56 · Major gaps | CLIFreemium | ★ 1.6k | today | TypeScript | Codex · Claude Code · Claude API |
| Lynkr | 62 · Some gaps | CLIFree + model costs | ★ 637 | 2d ago | JavaScript | Codex · Claude Code · OpenAI API |
How does FollowAgents rate this agent?
Why each dimension lost points
The README says the gateway sits between an agent and the LLM API, compacts history in the background, and offers API key configuration and optional Slack notifications. It does not explain permission scope, key storage, transmission or retention details, or external side effects, so these areas receive only partial credit. The install command pipes a downloaded script directly to a shell, and no recovery guidance is provided; rollback scores 0. The README associates the product with Compresr, but publisher identity remains unknown under the supplied provenance, and maintenance responsibility is unclear.
The Go module, build workflow, and README are broadly consistent about the gateway product. The workflow lists unit and integration tests and cross-platform builds, but integration tests require secrets and may be skipped, while vulnerability checks do not block CI; dependency availability therefore receives partial credit. The files provide no evidence about runtime error handling or failure messages, so failure_messages scores 0.
The README targets Claude Code, Cursor, OpenClaw, and custom configurations and gives a default 75% trigger threshold, covering several scenarios. It does not explain mode boundaries, trigger behavior in detail, or platform limitations. CI cross-platform builds provide limited environment evidence, but the installation instructions do not explain platform selection.
The README has a quick start, feature overview, and supported-agent list, giving it a reasonably clear structure. Installation relies on an online script and includes no troubleshooting or uninstall guidance. Naming is consistent between the README and Go module, but there is no FAQ, limitations section, or clear version policy. The LICENSE contains the full Apache 2.0 text and earns full marks. The CI workflows provide some evidence of maintenance activity, but name no responsible party or support commitment.
The README describes background summary precomputation intended to reduce waiting when context limits are reached and points users to compaction logs, indicating potential practical value. It provides no sample output, measured results, or cost information, so usability, marginal value, and cost-benefit receive only claim-level partial credit.
Product positioning and installation steps are traceable to the README, while the Go module and workflows provide limited corroboration about implementation and maintenance. There is no code, test result, or independent source corroborating product-effect or data-handling claims. The README does not distinguish demonstrated outcomes from expected results, so each criterion receives only partial credit.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- The installation step executes an online-fetched script through a shell; the supplied material does not describe script review, uninstall, or configuration recovery.
- The gateway handles conversation history, but the material does not say where summarization requests are sent, what logs contain, how keys are stored, or how long data is retained.
- CI integration tests may be skipped when secrets are absent, and vulnerability findings do not block CI; this does not establish dependency safety or that tests passed.
FAQ
Do I need to provide an API key?
Which agents can I configure?
When does compaction happen?
Where can I inspect compaction activity?
logs/history_compaction.jsonl. It does not describe the log fields or error handling.