floatboat
Open navigation menu

Context Mode

Context Mode keeps large tool outputs in a local SQLite index so a coding agent can retrieve the relevant lines instead of reading the entire result.

Make a noisy test log searchable

Design a local evidence workflow that returns the failing test and related lines while keeping the full log available.

Describe the command, log size and the errors you need to preserve.

Copy the request into a new Floatboat conversation and provide the required material.

Download Floatboat

Your request for Floatboat

Preview and copy the request. It is not sent automatically from this page.

Reference: Context Mode
https://github.com/mksglu/context-mode

Read the original documentation and check the current environment and dependencies. Propose any new installation or tool connection before proceeding.

My task:
Study Context Mode’s local indexing pattern. Given a test log, propose how to return failures by test name and retrieve supporting stack lines. Compare the evidence with the raw log and show how I can measure context usage.

Expected deliverable:
A log-indexing plan with retrieval examples and a check that the excerpts preserve the failure evidence.

How Context Mode handles large tool outputs

A full test log can fill an agent’s context while the useful evidence occupies only a few lines. Context Mode runs work in its local execution environment, keeps results in a SQLite FTS5 index and returns focused results or search excerpts.

MCP provides the tool interface. Host hooks and rules can steer compatible agents toward those tools, and continuity features vary by host. Read the platform guide for the agent you actually use and check what gets written to the local index before feeding it confidential material.

The local plugin is source-available under Elastic License 2.0. Context Mode Gateway is a separate hosted service. To borrow this pattern in Floatboat, keep raw evidence on disk and return compact references with enough context to verify a decision.

Context Mode’s execution and search tools

Long outputs can stay searchable even when only a few lines go back to the model. Review which tool outputs are indexed and how the agent retrieves details.

  • Local execution

    Run a command or script and return a focused result while preserving output for later lookup.

  • SQLite full-text search

    Retrieve passages from indexed tool results with an exact error string, test name or source identifier.

  • Host-specific routing

    Review MCP and hook support for your agent before expecting automatic routing or session recovery.

Context Mode use cases

Pick a noisy workflow and compare the evidence reaching the model. Track both the amount retrieved and whether it is enough to resolve the task.

Find failing tests in a long run

Return the failed test names and first errors to the agent. Search the saved output for an assertion and its surrounding stack when the first result is insufficient.

Review a batch of fetched pages

Index the fetch results under distinct source names. Retrieve only excerpts relevant to the question and preserve a link back to each original source.

Summarize a noisy build

Group errors by compiler message or file before handing them to the model. Keep the full log available to check whether different symptoms share a cause.

Resume a long investigation

On a host with continuity support, test whether the agent can recover previous decisions after compaction. Check the restored summary against a short reference before relying on it.

What to check before installing Context Mode

The local plugin and the hosted Gateway have separate setup and data flows. For a pilot, choose a supported host and use a test project whose outputs can be indexed safely.

  • Supported host

    Check which MCP tools, hooks and recovery features are available for your agent version.

  • Local data scope

    Review what the index retains, where it is stored and how to remove test data when the pilot ends.

  • License and deployment

    Review ELv2, especially the restrictions on offering the software as a managed service.

Related Software Projects

Compare the workflows, operating requirements, and source licenses of these projects.

Context Mode FAQ

Does Context Mode shorten the agent’s replies?
Its main optimization is on the input side: large tool results are kept out of the active context until requested.
Is the plugin open source?
The repository is source-available under Elastic License 2.0, which restricts certain uses such as offering its features as a managed service.
Is the local plugin the same as Gateway?
Gateway is a separate hosted service. Check its request routing and retention terms independently of the local plugin.
Do all agent hosts offer the same hooks?
Support varies by host. Check the platform documentation for routing, capture and session restoration.
Do the upstream byte benchmarks predict my token usage?
They describe particular workloads and units. Measure tokens or cost separately for the task, model and host you use.
Can I inspect an excerpt in context?
Search the stored output by a source name, error message or test identifier, then retrieve surrounding lines when the excerpt is too narrow.