Configuration

Eidet stores its configuration in a JSON file at:

Platform	Path
Windows	%APPDATA%\Eidet\eidet.json
macOS/Linux	~/.eidet/eidet.json

Managing Config

# View all settings
eidet config list

# Get a single value
eidet config get storage.ravenUrl

# Set a value
eidet config set service.port 19380

# JSON output
eidet config list --json

Environment Variable Overrides

Environment variables take precedence over the config file:

Variable	Config path	Example
EIDET_API_URL	service.bindAddress + service.port	http://0.0.0.0:19380
EIDET_RAVEN_URL	storage.ravenUrl	http://ravendb:8080
EIDET_OLLAMA_URL	enrichment.ollamaUrl	http://ollama:11434
EIDET_OLLAMA_MODEL	enrichment.ollamaModel	gemma4

These are useful for Docker and CI/CD environments.

Config Reference

service

Key	Type	Default	Description
service.port	int	19380	HTTP port for REST API and MCP
service.bindAddress	string	127.0.0.1	Bind address (use 0.0.0.0 for network access)

storage

Key	Type	Default	Description
storage.mode	enum	External	External or Embedded
storage.ravenUrl	string	http://localhost:8080	RavenDB connection URL
storage.databaseName	string	Eidet	RavenDB database name
storage.dataDir	string	null	Data directory for embedded mode

memory

Key	Type	Default	Description
memory.l1Count	int	20	Max memories in L1 context
memory.l1MaxTokens	int	500	Max tokens for L1 context
memory.duplicateThreshold	float	0.92	Vector similarity threshold for duplicate detection
memory.vectorSimilarityMinimum	float	0.70	Minimum similarity for vector search results
memory.observationRetentionDays	int	90	Days before old observations are cleaned up
memory.autoIntakeOnFirstSession	bool	true	Auto-ingest project files on first MCP session
memory.crossRepoRecallEnabled	bool	true	Include cross-repo results in recall
memory.stalenessWarningDays	int	7	Days before adding staleness warnings
memory.recallCacheEnabled	bool	true	Cache recall results (5-min TTL)

maintenance

Key	Type	Default	Description
maintenance.intervalHours	int	24	Hours between maintenance runs
maintenance.consolidationIntervalHours	int	6	Hours between consolidation runs
maintenance.atLocalTime	string	03:00	Local wall-clock time the nightly pass is anchored to

The anchor is what keeps the pass where you put it. Runs are scheduled on a grid anchored to atLocalTime, not at “whenever the last one finished plus intervalHours” — a pass that takes two hours would otherwise start two hours later every day and walk into the working day. A long or missed run costs its own slot and nothing more.

update

Key	Type	Default	Description
update.check	bool	true	Look for new releases once a night, and mention one when it exists
update.autoUpdate	bool	false	Install what it finds, without asking. eidet setup asks once
update.atLocalTime	string	04:00	Local wall-clock time for the nightly check
update.minimumAgeHours	int	24	Refuse to auto-install a release younger than this

Checking and installing are separate switches. With autoUpdate off you still get told a new version exists — once per process, on the CLI, in eidet_context, and as a banner in the Web UI — and install it yourself with eidet update.

Only the nightly task reaches NuGet; everything else reads its cached answer, so an MCP session start never pays a network round-trip. Nothing is checked at all with update.check set to false.

minimumAgeHours exists because releases are immutable: a bad build can only be superseded, never replaced, so waiting a day leaves room to publish the fix before the fleet takes the bad one. A release whose publish date can’t be read is treated as too young rather than old enough.

Auto-update applies only to dotnet tool installs. Container and standalone-binary installs are replaced as a whole, so they get the notice and nothing else.

eidet update --check          # ask now, print the answer, install nothing
eidet update                  # install the newest release
eidet update --to 0.11.2      # install one specific version
eidet update --rollback       # go back to the previously installed version

enrichment

Key	Type	Default	Description
enrichment.ollamaEnabled	bool	false	Enable Ollama enrichment
enrichment.ollamaUrl	string	http://localhost:11434	Ollama API URL
enrichment.ollamaModel	string	gemma4	Model for enrichment tasks
enrichment.apiKey	string	—	Bearer token sent with every request. For a private network cluster behind a gateway; local servers ignore it. config get/list print (set), never the value
enrichment.thinking	bool	unset	false turns a reasoning model’s thinking off (chat_template_kwargs.thinking on vLLM/llama.cpp, think on Ollama) — ~5x cheaper per call. Unset sends nothing; default unsets
enrichment.fallbacks	array	[]	Backends tried in order when the one before is offline or fails a call — each with provider, url, model, optional apiKey, thinking. Edit in the file or via eidet enrichment setup
enrichment.autoOneLiner	bool	true	Auto-generate one-liner summaries
enrichment.autoForesight	bool	true	Auto-generate foresight hints
enrichment.autoConsolidation	bool	true	Use LLM for consolidation merges
enrichment.driftReview.enabled	bool	true	Nightly LLM re-read of stored memories for staleness
enrichment.driftReview.nightlyBatch	int	25	Model calls per repo per run — the recurring cost
enrichment.driftReview.minAgeDays	int	7	Ignore memories younger than this
enrichment.driftReview.reviewIntervalDays	int	90	How long a verdict stands before re-review. 0 = every night
enrichment.driftReview.minModelConfidence	float	0.7	Below this a verdict is recorded but not acted on
enrichment.driftReview.autonomy	string	Decay	FlagOnly, Decay, or Expire
enrichment.reflection.enabled	bool	false	Mint new memories from feedback residue (dormant)

Drift review is the only enrichment surface whose cost recurs: it is nightlyBatch model calls per repo per night. reviewIntervalDays is what makes it converge — a memory drops out of the candidate set until its verdict ages past the interval, so a corpus nobody is touching costs nothing. Setting it to 0 restores an unbounded nightly sweep. The startup banner’s Nightly AI: line reports what the running service will actually spend, and so does eidet enrichment reload, which applies drift-review and reflection changes to the running service without a restart. Every enrichment.* key in this table is an eidet config set key.

A private network model in front of a local one is the intended shape for fallbacks: the network cluster is fast but not always reachable, the local server always is. Each backend keeps its own health verdict (cached five minutes, probed with a five-second budget), so an offline primary costs one probe per five minutes and every call goes straight to the fallback. eidet doctor shows one row per backend; the service log names the backend that answers and reports when a fallback takes over. The URL may be written with or without a trailing /v1 — both mean the same server.

{
  "enrichment": {
    "enabled": true,
    "provider": "OpenAiCompatible",
    "url": "https://cortex.example.com",
    "model": "deepseek-v4-flash",
    "apiKey": "…",
    "thinking": false,
    "fallbacks": [
      { "provider": "Ollama", "url": "http://localhost:11434", "model": "gemma4" }
    ]
  }
}

auth

Key	Type	Default	Description
auth.enabled	bool	false	Require API key authentication
auth.requireForNonLocalhost	bool	true	Block non-localhost without auth

Manage API keys:

eidet api-key create "my-key" --scopes read:all,write:all
eidet api-key list
eidet api-key revoke <key-id>

Scopes: read:all, write:observations, write:all, admin (implies all).

hooks

Hooks are configured in the config file directly (not via config set). See Hooks for details.

{
  "hooks": {
    "preStore": [
      { "command": "python validate.py", "timeoutSeconds": 10, "enabled": true }
    ],
    "postStore": [],
    "preRecall": [],
    "postRecall": [],
    "preForget": [],
    "postForget": []
  }
}

Example Config File

{
  "service": {
    "port": 19380,
    "bindAddress": "127.0.0.1"
  },
  "storage": {
    "mode": "Embedded",
    "ravenUrl": "http://localhost:8080",
    "databaseName": "Eidet"
  },
  "memory": {
    "l1Count": 20,
    "duplicateThreshold": 0.92,
    "autoIntakeOnFirstSession": true
  },
  "update": {
    "check": true,
    "autoUpdate": false,
    "atLocalTime": "04:00",
    "minimumAgeHours": 24
  },
  "enrichment": {
    "ollamaEnabled": true,
    "ollamaModel": "gemma4"
  },
  "auth": {
    "enabled": false
  },
  "hooks": {
    "preStore": [],
    "postStore": []
  }
}